Skip to content

Integrate fused SwiGLU decode into Llama 3.2 1B#80

Draft
jgmelber wants to merge 11 commits intodevelfrom
decode-fusion-llama
Draft

Integrate fused SwiGLU decode into Llama 3.2 1B#80
jgmelber wants to merge 11 commits intodevelfrom
decode-fusion-llama

Conversation

@jgmelber
Copy link
Collaborator

@jgmelber jgmelber commented Mar 7, 2026

Summary

  • Integrates AIESwiGLUFusedDecode operator into the Llama 3.2 1B application
  • Adds use_aie_ffn_swiglu_fused config flag (enabled by default in llama32_1b.json)
  • When enabled, replaces the 2-runlist AIESwiGLUDecode with the 1-runlist fused operator (1.32x faster MLP at production dims)
  • Skips creating redundant decode GEMV operators (aie_fc1/fc2/fc3_gemv) when fused path handles everything

Test plan

  • CI linting passes (black, clang-format, reuse)
  • CI small test suite passes
  • CI application tests pass with fused config
  • Verify Llama inference produces correct output with fused SwiGLU

🤖 Generated with Claude Code

jgmelber and others added 11 commits March 7, 2026 10:13
Eliminates a full DRAM round-trip for the intermediate left_swished
buffer in the SwiGLU pipeline by computing silu(x) * y in a single
vectorized kernel loop. Reduces SwiGLU prefill from 5 to 4 runlist
entries.

New operator: AIESiLUMul with fused C++ kernels for both AIE2 (LUT tanh)
and AIE2+ (hardware tanh). Integrated into swiglu_prefill.
Also fixes a pre-existing bug in swiglu_prefill/test.py (errors_2 -> errors_3).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Collapses three separate NPU designs (GEMV W1, GEMV W2, SiLU+Mul) into
a single fused operator. Each AIE core loads vector x once, processes
both W1 and W2 rows through a shared A FIFO with pre-interleaved weights,
then computes silu(left)*right entirely in L1 via kernel-local static
buffers. The intermediate vectors never touch DRAM.

Reduces SwiGLU decode from 4 to 2 runlist entries and eliminates the
left/right buffer allocations. Uses 4 AIE columns (DMA channel limit:
2 input + 1 output per tile).

Note: swiglu_prefill unchanged — uses GEMM (not GEMV) so dual-GEMV
fusion does not apply.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…on, INT4 dequant-GEMV

Implements Phases 1-3 and 5 from the decode dataflow optimization plan:

Phase 1 - fused_qkv_proj: Concatenates Wq+Wk+Wv into single GEMV (M=3072,
K=2048), eliminating 3 redundant input vector loads. Reuses existing mv.o.

Phase 2 - flowkv_decode: Streaming decode attention with online softmax.
2-tile pipeline per KV head group (score tile + value tile). Intermediates
flow tile-to-tile via on-chip ObjectFIFOs. Uses aie::exp2 for safe exp.

Phase 3 - swiglu_fused_decode: Complete SwiGLU fusion with 2-stage tile
pipeline. Dual-GEMV+SiLU+Mul feeds directly into down-projection GEMV via
inter-tile ObjectFIFO. The 32 KB intermediate never touches DDR. Host
reduces 4 column partials. Benchmarked 1.32x speedup at Llama dims
(5410 us -> 4103 us, 24.5 GB/s effective bandwidth).

Phase 5 - fused_dequant_gemv: Fused INT4 weight dequantization + GEMV in
single kernel pass. Uses proven aie::unpack chain from expand.cc for
INT4->bf16 conversion. 4x DDR bandwidth reduction.

Also fixes:
- Static buffer overflow in dual_gemv_silu_mul (1024 -> 2048 elements)
- Production-scale test params for swiglu_decode (2048, 8192)

All operators verified on AMD Ryzen AI 9 HX 370 (RyzenAI-npu4) hardware.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
swiglu_fused_decode: Replace DDR partial-sum reduction with on-chip
MemTile join + dedicated reduction tile. Stage 2 partials flow through
ObjectFIFO join into a single reduction tile that sums element-wise,
producing the final output directly. Eliminates 12 KB of DDR output
traffic (16 KB partials -> 4 KB single output) and removes host-side
partials.sum(dim=0) call. 3-stage pipeline: dual-GEMV+SiLU+Mul ->
down-proj GEMV -> on-chip reduce.

flowkv_decode: Fuse RoPE rotation into the score tile kernel. Q angles
(cos/sin interleaved, 128 bytes) are packed into the Q FIFO buffer
alongside query vectors, staying within the 2-input DMA channel limit.
The score tile applies two-halves RoPE to Q in-register before the
Q*K^T dot product. K is assumed pre-rotated in the KV cache (standard
practice). Eliminates RoPE as a separate operator invocation.

Both changes verified on AMD Ryzen AI NPU hardware (10/10 tests passed).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
On-chip MemTile reduction added ~400 us pipeline serialization to save
only 12 KB DDR traffic (~0.5 us). DDR partials with host sum() is 1.32x
faster than baseline vs 1.20x with on-chip reduce. Keep the faster
approach.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add use_aie_ffn_swiglu_fused config flag that swaps AIESwiGLUDecode
(2-runlist, DDR intermediate) for AIESwiGLUFusedDecode (1-runlist,
on-chip intermediate) during decode. Provides 1.32x MLP speedup at
Llama production dimensions.

Enabled via JSON config: "use_aie_ffn_swiglu_fused": true
(requires use_aie_ffn_swiglu: true and use_kv_cache: true)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
When use_aie_ffn_swiglu_fused is True, the fused operator handles all
of gate+up+silu+mul+down in one design. Skip creating the 3 separate
decode GEMVs (aie_fc1_gemv, aie_fc2_gemv, aie_fc3_gemv) which would
waste compilation time and device memory.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
@jgmelber jgmelber changed the title Integrate fused SwiGLU decode into Llama 3.2 1B Integrate fused decode into Llama 3.2 1B Mar 7, 2026
@jgmelber jgmelber changed the title Integrate fused decode into Llama 3.2 1B Integrate fused SwiGLU decode into Llama 3.2 1B Mar 7, 2026
@github-actions
Copy link
Contributor

github-actions bot commented Mar 7, 2026

📊 Test Results for Small Benchmark/Test Suite

116b247 (2026_03_07_19_09_31)

IRONCLAD

Tested on 2026_03_07_19_09_31 at commit 116b247.

Test Checks Latency (mean)Bandwidth (mean)GFLOP/s (mean)
axpy_1_cols_2_channels_2048_tile_2048_3.0_0 ✅ 5/5 113.82 0.12 n/a
axpy_2_cols_2_channels_2048_tile_1024_3.0_0 ✅ 5/5 107.18 0.12 n/a
axpy_4_cols_2_channels_2048_tile_512_3.0_0 ✅ 5/5 99.78 0.12 n/a
axpy_8_cols_2_channels_2048_tile_256_3.0_0 ✅ 5/5 125.62 0.10 n/a
dequant_1_cols_1_channels_2048_tile_2048_0 ✅ 5/5 118.66 0.05 n/a
dequant_1_cols_2_channels_2048_tile_1024_0 ✅ 5/5 102.94 0.06 n/a
dequant_2_cols_1_channels_2048_tile_1024_0 ✅ 5/5 114.42 0.05 n/a
dequant_2_cols_2_channels_2048_tile_512_0 ✅ 5/5 128.62 0.04 n/a
dequant_4_cols_1_channels_2048_tile_512_0 ✅ 5/5 108.48 0.05 n/a
dequant_4_cols_2_channels_2048_tile_256_0 ✅ 5/5 123.18 0.04 n/a
dequant_8_cols_1_channels_2048_tile_256_0 ✅ 5/5 125.24 0.04 n/a
dequant_8_cols_2_channels_2048_tile_128_0 ✅ 5/5 148.24 0.04 n/a
dual_gemv_silu_mul_2048x2048_4tsi_512tso_4col0 ✅ 5/5 461.52 36.42 n/a
eltwise_add_1_cols_2_channels_2048_tile_2048 ✅ 5/5 112.82 0.11 n/a
eltwise_add_2_cols_2_channels_2048_tile_1024 ✅ 5/5 118.50 0.11 n/a
eltwise_add_4_cols_2_channels_2048_tile_512 ✅ 5/5 118.46 0.11 n/a
eltwise_add_8_cols_2_channels_2048_tile_256 ✅ 5/5 102.94 0.12 n/a
eltwise_mul_1_cols_2_channels_2048_tile_2048 ✅ 5/5 121.06 0.10 n/a
eltwise_mul_2_cols_2_channels_2048_tile_1024 ✅ 5/5 119.06 0.11 n/a
eltwise_mul_4_cols_2_channels_2048_tile_512 ✅ 5/5 86.92 0.15 n/a
eltwise_mul_8_cols_2_channels_2048_tile_256 ✅ 5/5 103.78 0.12 n/a
flowkv_decode_32h_8kv_64d_128s_32cs_4col0 ✅ 5/5 348.94 0.78 n/a
fused_dequant_gemv_2048x2048_1tsi_512tso_4col_g32_0 ✅ 5/5 1260.80 1.88 n/a
fused_qkv_proj_3072x2048_4tsi_768tso_4col0 ✅ 5/5 371.78 33.92 n/a
gelu_1_cols_1_channels_2048_tile_2048 ✅ 5/5 102.14 0.09 n/a
gelu_1_cols_2_channels_2048_tile_1024 ✅ 5/5 94.12 0.09 n/a
gelu_2_cols_1_channels_2048_tile_1024 ✅ 5/5 100.46 0.09 n/a
gelu_2_cols_2_channels_2048_tile_512 ✅ 5/5 88.72 0.10 n/a
gelu_4_cols_1_channels_2048_tile_512 ✅ 5/5 100.68 0.09 n/a
gelu_4_cols_2_channels_2048_tile_256 ✅ 5/5 113.10 0.08 n/a
gelu_8_cols_1_channels_2048_tile_256 ✅ 5/5 102.84 0.08 n/a
gelu_8_cols_2_channels_2048_tile_128 ✅ 5/5 117.66 0.07 n/a
gemm_1792x896x1152_64x32x48_8cols_ccolmaj ✅ 5/5 1915.48 4.92 n/a
gemm_192x384x64_48x96x16_4cols ✅ 5/5 138.56 1.68 n/a
gemm_192x384x64_48x96x16_4cols_bcolmaj_ccolmaj ✅ 5/5 139.70 1.70 n/a
gemm_2048x2048x2048_64x64x64_1cols ✅ 5/5 48422.36 0.52 n/a
gemm_2048x2048x2048_64x64x64_2cols_bcolmaj ✅ 5/5 27666.40 0.91 n/a
gemm_2048x2048x2048_64x64x64_8cols_bcolmaj_ccolmaj ✅ 5/5 7618.82 3.31 n/a
gemm_384x1536x1792_32x48x64_4cols_bcolmaj ✅ 5/5 1925.26 4.20 n/a
gemm_896x1792x640_32x64x80_8cols_ccolmaj ✅ 5/5 1031.94 6.53 n/a
layer_norm_1_cols_1_channels_2048_tile_2048 ✅ 5/5 117.06 0.08 n/a
layer_norm_1_cols_2_channels_2048_tile_1024 ✅ 5/5 95.74 0.09 n/a
layer_norm_2_cols_1_channels_2048_tile_1024 ✅ 5/5 87.44 0.10 n/a
layer_norm_2_cols_2_channels_2048_tile_512 ✅ 5/5 88.54 0.10 n/a
layer_norm_4_cols_1_channels_2048_tile_512 ✅ 5/5 90.72 0.10 n/a
layer_norm_4_cols_2_channels_2048_tile_256 ✅ 5/5 125.50 0.07 n/a
layer_norm_8_cols_1_channels_2048_tile_256 ✅ 5/5 108.34 0.08 n/a
layer_norm_8_cols_2_channels_2048_tile_128 ✅ 5/5 130.96 0.07 n/a
matrix_vector_mul_128x128_32tsi_128tso_1col0 ✅ 5/5 n/a 0.31 n/a
matrix_vector_mul_2048x8192_1tsi_1024tso_2col0 ✅ 5/5 n/a 23.14 n/a
matrix_vector_mul_2048x8192_1tsi_2048tso_1col0 ✅ 5/5 n/a 13.09 n/a
matrix_vector_mul_2048x8192_1tsi_256tso_8col0 ✅ 5/5 n/a 44.80 n/a
matrix_vector_mul_2048x8192_1tsi_512tso_4col0 ✅ 5/5 n/a 39.87 n/a
matrix_vector_mul_8192x2048_4tsi_1024tso_1col0 ✅ 5/5 n/a 13.19 n/a
matrix_vector_mul_8192x2048_4tsi_1024tso_2col0 ✅ 5/5 n/a 24.28 n/a
matrix_vector_mul_8192x2048_4tsi_1024tso_4col0 ✅ 5/5 n/a 36.64 n/a
matrix_vector_mul_8192x2048_4tsi_1024tso_8col0 ✅ 5/5 n/a 46.71 n/a
mem_copy_16_cores_2_chans_2048_tile_128_False0 ✅ 5/5 116.16 0.07 n/a
mem_copy_1_cores_1_chans_2048_tile_2048_False0 ✅ 5/5 87.96 0.09 n/a
mem_copy_2_cores_1_chans_2048_tile_1024_False0 ✅ 5/5 108.20 0.08 n/a
mem_copy_2_cores_2_chans_2048_tile_1024_False0 ✅ 5/5 101.62 0.08 n/a
mem_copy_4_cores_1_chans_2048_tile_512_False0 ✅ 5/5 126.36 0.07 n/a
mem_copy_4_cores_2_chans_2048_tile_512_False0 ✅ 5/5 96.78 0.09 n/a
mem_copy_8_cores_1_chans_2048_tile_256_False0 ✅ 5/5 127.80 0.07 n/a
mem_copy_8_cores_2_chans_2048_tile_256_False0 ✅ 5/5 112.54 0.07 n/a
mha_16384_64_1_8_0_0 ✅ 5/5 47329.26 0.18 n/a
relu_1_cols_1_channels_2048_tile_2048 ✅ 5/5 111.32 0.08 n/a
relu_2_cols_1_channels_2048_tile_1024 ✅ 5/5 114.20 0.08 n/a
relu_4_cols_1_channels_2048_tile_512 ✅ 5/5 114.68 0.08 n/a
relu_8_cols_1_channels_2048_tile_256 ✅ 5/5 116.60 0.07 n/a
rms_norm_1_cols_1_channels_2048_tile_2048 ✅ 5/5 107.22 0.08 n/a
rms_norm_1_cols_2_channels_2048_tile_1024 ✅ 5/5 115.84 0.08 n/a
rms_norm_2_cols_1_channels_2048_tile_1024 ✅ 5/5 100.48 0.08 n/a
rms_norm_2_cols_2_channels_2048_tile_512 ✅ 5/5 112.96 0.08 n/a
rms_norm_4_cols_1_channels_2048_tile_512 ✅ 5/5 131.30 0.07 n/a
rms_norm_4_cols_2_channels_2048_tile_256 ✅ 5/5 99.70 0.09 n/a
rms_norm_8_cols_1_channels_2048_tile_256 ✅ 5/5 122.44 0.07 n/a
rms_norm_8_cols_2_channels_2048_tile_128 ✅ 5/5 99.12 0.09 n/a
rope_1c_32rows_512cols_32arows_0m ✅ 5/5 108.96 0.96 n/a
rope_1c_32rows_512cols_8arows_0m ✅ 5/5 156.58 0.47 n/a
rope_2c_32rows_512cols_32arows_0m ✅ 5/5 108.36 0.99 n/a
rope_2c_32rows_512cols_8arows_0m ✅ 5/5 120.16 0.63 n/a
rope_8c_32rows_512cols_32arows_0m ✅ 5/5 103.04 1.01 n/a
rope_8c_32rows_512cols_8arows_0m ✅ 5/5 145.26 0.54 n/a
sigmoid_1_cols_1_channels_2048_tile_2048 ✅ 5/5 118.84 0.07 n/a
sigmoid_2_cols_1_channels_2048_tile_1024 ✅ 5/5 108.54 0.08 n/a
sigmoid_4_cols_1_channels_2048_tile_512 ✅ 5/5 105.56 0.08 n/a
sigmoid_8_cols_1_channels_2048_tile_256 ✅ 5/5 122.40 0.07 n/a
silu_1_cols_1_channels_2048_tile_2048 ✅ 5/5 117.24 0.07 n/a
silu_2_cols_1_channels_2048_tile_1024 ✅ 5/5 111.16 0.08 n/a
silu_4_cols_1_channels_2048_tile_512 ✅ 5/5 109.12 0.08 n/a
silu_8_cols_1_channels_2048_tile_256 ✅ 5/5 133.34 0.07 n/a
silu_mul_1_cols_2_channels_2048_tile_2048 ✅ 5/5 100.84 0.13 n/a
silu_mul_2_cols_2_channels_2048_tile_1024 ✅ 5/5 112.68 0.11 n/a
silu_mul_4_cols_2_channels_2048_tile_512 ✅ 5/5 95.12 0.13 n/a
silu_mul_8_cols_2_channels_2048_tile_256 ✅ 5/5 102.32 0.13 n/a
softmax_1_cols_2_channels_4096_tile_2048 ✅ 5/5 102.30 0.17 n/a
softmax_2_cols_2_channels_4096_tile_1024 ✅ 5/5 114.22 0.16 n/a
softmax_2_cols_2_channels_4096_tile_512 ✅ 5/5 100.84 0.17 n/a
swiglu_decode_1x2048x2048_0 ✅ 5/5 1574.80 0.00 n/a
swiglu_fused_decode_2048x2048_0 ✅ 5/5 668.98 0.01 n/a
tanh_1_cols_1_channels_2048_tile_2048 ✅ 5/5 124.20 0.07 n/a
tanh_2_cols_1_channels_2048_tile_1024 ✅ 5/5 111.16 0.08 n/a
tanh_4_cols_1_channels_2048_tile_512 ✅ 5/5 115.48 0.07 n/a
tanh_8_cols_1_channels_2048_tile_256 ✅ 5/5 117.10 0.07 n/a
transpose_2048_M_64_N_1_cols_1_channels_64_m_64_n_8_s0 ✅ 5/5 151.18 3.54 n/a
transpose_2048_M_64_N_1_cols_2_channels_64_m_64_n_8_s0 ✅ 5/5 134.52 4.14 n/a
weighted_rms_norm_1_cols_2_channels_2048_weights_2048 ✅ 5/5 112.18 0.08 n/a
weighted_rms_norm_2_cols_2_channels_2048_weights_1024 ✅ 5/5 113.70 0.08 n/a
weighted_rms_norm_4_cols_2_channels_2048_weights_512 ✅ 5/5 101.50 0.09 n/a
weighted_rms_norm_8_cols_2_channels_2048_weights_256 ✅ 5/5 122.86 0.07 n/a
📈 Trends (vs main branch) for Small Benchmark/Test Suite

116b247 (2026_03_07_19_09_31)

IRONCLAD Trends

axpy_1_cols_2_channels_2048_tile_2048_3.0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
130b6ea — 2025-12-05 21:15:490.13 (-1.96%)0.11 (+2.52%)0.10 (+1.86%)0.08 (-1.91%)0.02 (-1.72%)162.40 (+1.95%)120.96 (-2.36%)117.00 (-1.85%)94.40 (+2.05%)23.11 (+4.09%)
0a6c11c — 2025-12-04 00:39:100.13 (n/a)0.10 (n/a)0.10 (n/a)0.08 (n/a)0.02 (n/a)159.30 (n/a)123.88 (n/a)119.20 (n/a)92.50 (n/a)22.21 (n/a)

axpy_1_cols_2_channels_2048_tile_2048_3.0_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.16 (-0.01%)0.12 (-14.38%)0.12 (-15.87%)0.07 (-27.35%)0.04 (+19.91%)173.00 (+37.63%)113.82 (+21.65%)99.20 (+18.80%)75.00 (+0.00%)38.50 (+72.02%)
897d04e — 2026-03-06 22:12:250.16 (n/a)0.14 (n/a)0.15 (n/a)0.10 (n/a)0.03 (n/a)125.70 (n/a)93.56 (n/a)83.50 (n/a)75.00 (n/a)22.38 (n/a)

axpy_2_cols_2_channels_2048_tile_1024_3.0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
130b6ea — 2025-12-05 21:15:490.11 (-26.77%)0.09 (-9.09%)0.09 (-10.81%)0.08 (+8.25%)0.01 (-55.42%)152.00 (-7.60%)135.46 (+6.28%)142.80 (+12.18%)115.00 (+36.58%)15.03 (-41.36%)
0a6c11c — 2025-12-04 00:39:100.15 (n/a)0.10 (n/a)0.10 (n/a)0.07 (n/a)0.02 (n/a)164.50 (n/a)127.46 (n/a)127.30 (n/a)84.20 (n/a)25.64 (n/a)

axpy_2_cols_2_channels_2048_tile_1024_3.0_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.14 (+10.38%)0.12 (+8.80%)0.12 (+5.57%)0.10 (+43.44%)0.02 (-28.82%)127.40 (-30.27%)107.18 (-11.32%)105.00 (-5.23%)87.00 (-9.47%)15.03 (-57.25%)
897d04e — 2026-03-06 22:12:250.13 (n/a)0.11 (n/a)0.11 (n/a)0.07 (n/a)0.02 (n/a)182.70 (n/a)120.86 (n/a)110.80 (n/a)96.10 (n/a)35.16 (n/a)

axpy_4_cols_2_channels_2048_tile_512_3.0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
130b6ea — 2025-12-05 21:15:490.12 (-10.21%)0.10 (-8.41%)0.10 (-6.20%)0.07 (+3.52%)0.02 (-23.48%)166.70 (-3.36%)128.78 (+7.33%)121.80 (+6.65%)103.90 (+11.36%)22.95 (-18.82%)
0a6c11c — 2025-12-04 00:39:100.13 (n/a)0.11 (n/a)0.11 (n/a)0.07 (n/a)0.02 (n/a)172.50 (n/a)119.98 (n/a)114.20 (n/a)93.30 (n/a)28.27 (n/a)

axpy_4_cols_2_channels_2048_tile_512_3.0_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.13 (+4.84%)0.12 (+17.29%)0.12 (+1.16%)0.12 (+86.20%)0.01 (-75.14%)104.30 (-46.26%)99.78 (-19.87%)102.10 (-1.16%)92.80 (-4.53%)5.03 (-87.52%)
897d04e — 2026-03-06 22:12:250.13 (n/a)0.11 (n/a)0.12 (n/a)0.06 (n/a)0.03 (n/a)194.10 (n/a)124.52 (n/a)103.30 (n/a)97.20 (n/a)40.32 (n/a)

axpy_8_cols_2_channels_2048_tile_256_3.0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
130b6ea — 2025-12-05 21:15:490.10 (-16.19%)0.08 (-10.40%)0.09 (-2.09%)0.07 (-17.38%)0.01 (-10.61%)184.40 (+21.00%)153.18 (+11.99%)144.10 (+2.20%)127.10 (+19.34%)21.26 (+34.76%)
0a6c11c — 2025-12-04 00:39:100.12 (n/a)0.09 (n/a)0.09 (n/a)0.08 (n/a)0.01 (n/a)152.40 (n/a)136.78 (n/a)141.00 (n/a)106.50 (n/a)15.78 (n/a)

axpy_8_cols_2_channels_2048_tile_256_3.0_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.12 (-7.67%)0.10 (-8.87%)0.11 (-1.15%)0.08 (+0.94%)0.02 (-14.63%)154.80 (-0.96%)125.62 (+8.99%)114.00 (+1.15%)103.40 (+8.27%)22.25 (-8.56%)
897d04e — 2026-03-06 22:12:250.13 (n/a)0.11 (n/a)0.11 (n/a)0.08 (n/a)0.02 (n/a)156.30 (n/a)115.26 (n/a)112.70 (n/a)95.50 (n/a)24.34 (n/a)

dequant_1_cols_1_channels_2048_tile_2048

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
130b6ea — 2025-12-05 21:15:490.05 (-18.83%)0.04 (-5.21%)0.04 (-2.69%)0.03 (+37.10%)0.00 (-53.98%)160.40 (-27.06%)141.12 (-0.91%)136.40 (+2.79%)114.30 (+23.17%)17.37 (-59.06%)
0a6c11c — 2025-12-04 00:39:100.06 (n/a)0.04 (n/a)0.04 (n/a)0.02 (n/a)0.01 (n/a)219.90 (n/a)142.42 (n/a)132.70 (n/a)92.80 (n/a)42.43 (n/a)

dequant_1_cols_1_channels_2048_tile_2048_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.06 (+19.63%)0.05 (+9.87%)0.04 (-1.48%)0.04 (+8.11%)0.01 (+71.91%)146.60 (-7.45%)118.66 (-7.38%)125.60 (+1.45%)92.30 (-16.39%)23.30 (+26.93%)
897d04e — 2026-03-06 22:12:250.05 (n/a)0.04 (n/a)0.04 (n/a)0.03 (n/a)0.01 (n/a)158.40 (n/a)128.12 (n/a)123.80 (n/a)110.40 (n/a)18.36 (n/a)

dequant_1_cols_2_channels_2048_tile_1024

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
130b6ea — 2025-12-05 21:15:490.06 (-13.28%)0.05 (-8.93%)0.06 (-13.59%)0.04 (+4.63%)0.01 (-39.10%)133.60 (-4.43%)106.08 (+5.85%)95.10 (+15.69%)87.90 (+15.35%)17.90 (-31.97%)
0a6c11c — 2025-12-04 00:39:100.07 (n/a)0.06 (n/a)0.06 (n/a)0.04 (n/a)0.01 (n/a)139.80 (n/a)100.22 (n/a)82.20 (n/a)76.20 (n/a)26.31 (n/a)

dequant_1_cols_2_channels_2048_tile_1024_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.07 (+2.39%)0.06 (+11.22%)0.06 (+19.74%)0.03 (-5.24%)0.02 (+28.32%)171.60 (+5.47%)102.94 (-7.13%)82.50 (-16.50%)79.60 (-2.33%)39.40 (+27.04%)
897d04e — 2026-03-06 22:12:250.06 (n/a)0.05 (n/a)0.05 (n/a)0.03 (n/a)0.01 (n/a)162.70 (n/a)110.84 (n/a)98.80 (n/a)81.50 (n/a)31.02 (n/a)

dequant_2_cols_1_channels_2048_tile_1024

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
130b6ea — 2025-12-05 21:15:490.07 (+5.66%)0.05 (+1.21%)0.05 (+0.61%)0.04 (-15.02%)0.01 (+58.42%)143.40 (+17.64%)101.02 (+1.49%)97.60 (-0.61%)78.90 (-5.40%)22.56 (+78.52%)
0a6c11c — 2025-12-04 00:39:100.06 (n/a)0.05 (n/a)0.05 (n/a)0.04 (n/a)0.01 (n/a)121.90 (n/a)99.54 (n/a)98.20 (n/a)83.40 (n/a)12.64 (n/a)

dequant_2_cols_1_channels_2048_tile_1024_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.06 (-7.46%)0.05 (+0.02%)0.05 (-1.39%)0.03 (-5.57%)0.01 (-13.33%)161.30 (+5.91%)114.42 (-0.57%)102.20 (+1.39%)95.20 (+8.06%)27.19 (-0.03%)
897d04e — 2026-03-06 22:12:250.06 (n/a)0.05 (n/a)0.05 (n/a)0.03 (n/a)0.01 (n/a)152.30 (n/a)115.08 (n/a)100.80 (n/a)88.10 (n/a)27.20 (n/a)

dequant_2_cols_2_channels_2048_tile_512

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
130b6ea — 2025-12-05 21:15:490.05 (+17.06%)0.04 (+8.98%)0.05 (+21.37%)0.03 (-11.65%)0.01 (+301.89%)159.70 (+13.18%)126.32 (-5.68%)109.00 (-17.67%)106.80 (-14.56%)22.48 (+280.15%)
0a6c11c — 2025-12-04 00:39:100.04 (n/a)0.04 (n/a)0.04 (n/a)0.04 (n/a)0.00 (n/a)141.10 (n/a)133.92 (n/a)132.40 (n/a)125.00 (n/a)5.91 (n/a)

dequant_2_cols_2_channels_2048_tile_512_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.06 (-16.41%)0.04 (-6.61%)0.04 (-3.46%)0.03 (-14.16%)0.01 (-11.38%)191.90 (+16.51%)128.62 (+7.98%)124.50 (+3.58%)90.40 (+19.58%)40.16 (+27.05%)
897d04e — 2026-03-06 22:12:250.07 (n/a)0.05 (n/a)0.04 (n/a)0.03 (n/a)0.01 (n/a)164.70 (n/a)119.12 (n/a)120.20 (n/a)75.60 (n/a)31.61 (n/a)

dequant_4_cols_1_channels_2048_tile_512

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
130b6ea — 2025-12-05 21:15:490.06 (+20.24%)0.05 (-3.19%)0.05 (-6.73%)0.03 (-20.69%)0.01 (+194.26%)156.10 (+26.09%)120.86 (+7.15%)114.90 (+7.18%)88.30 (-16.85%)24.40 (+214.82%)
0a6c11c — 2025-12-04 00:39:100.05 (n/a)0.05 (n/a)0.05 (n/a)0.04 (n/a)0.00 (n/a)123.80 (n/a)112.80 (n/a)107.20 (n/a)106.20 (n/a)7.75 (n/a)

dequant_4_cols_1_channels_2048_tile_512_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.06 (+2.46%)0.05 (+11.79%)0.05 (+17.85%)0.04 (+17.25%)0.01 (-2.68%)140.40 (-14.70%)108.48 (-11.18%)98.30 (-15.19%)95.10 (-2.46%)19.33 (-22.93%)
897d04e — 2026-03-06 22:12:250.05 (n/a)0.04 (n/a)0.05 (n/a)0.03 (n/a)0.01 (n/a)164.60 (n/a)122.14 (n/a)115.90 (n/a)97.50 (n/a)25.08 (n/a)

dequant_4_cols_2_channels_2048_tile_256

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
130b6ea — 2025-12-05 21:15:490.05 (-6.09%)0.04 (-8.04%)0.04 (-12.46%)0.03 (-0.41%)0.01 (-20.19%)168.10 (+0.42%)135.74 (+7.44%)132.20 (+14.26%)107.40 (+6.44%)22.91 (-12.67%)
0a6c11c — 2025-12-04 00:39:100.05 (n/a)0.04 (n/a)0.05 (n/a)0.03 (n/a)0.01 (n/a)167.40 (n/a)126.34 (n/a)115.70 (n/a)100.90 (n/a)26.23 (n/a)

dequant_4_cols_2_channels_2048_tile_256_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.05 (-17.71%)0.04 (-2.61%)0.05 (+19.33%)0.03 (+8.51%)0.01 (-27.70%)153.30 (-7.87%)123.18 (+0.69%)107.00 (-16.21%)102.80 (+21.51%)24.94 (-18.16%)
897d04e — 2026-03-06 22:12:250.06 (n/a)0.05 (n/a)0.04 (n/a)0.03 (n/a)0.01 (n/a)166.40 (n/a)122.34 (n/a)127.70 (n/a)84.60 (n/a)30.47 (n/a)

dequant_8_cols_1_channels_2048_tile_256

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
130b6ea — 2025-12-05 21:15:490.04 (-13.65%)0.04 (-13.67%)0.04 (-18.56%)0.03 (-13.52%)0.00 (-23.88%)170.40 (+15.68%)140.12 (+15.33%)135.80 (+22.78%)119.60 (+15.78%)17.94 (+1.87%)
0a6c11c — 2025-12-04 00:39:100.05 (n/a)0.04 (n/a)0.05 (n/a)0.04 (n/a)0.01 (n/a)147.30 (n/a)121.50 (n/a)110.60 (n/a)103.30 (n/a)17.61 (n/a)

dequant_8_cols_1_channels_2048_tile_256_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.05 (+7.21%)0.04 (+2.13%)0.04 (+4.67%)0.03 (-9.93%)0.01 (+71.51%)158.40 (+11.00%)125.24 (-0.95%)120.10 (-4.46%)107.70 (-6.75%)19.42 (+83.08%)
897d04e — 2026-03-06 22:12:250.05 (n/a)0.04 (n/a)0.04 (n/a)0.04 (n/a)0.00 (n/a)142.70 (n/a)126.44 (n/a)125.70 (n/a)115.50 (n/a)10.61 (n/a)

dequant_8_cols_2_channels_2048_tile_128

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
130b6ea — 2025-12-05 21:15:490.05 (+17.21%)0.03 (-1.70%)0.03 (-3.04%)0.03 (-16.20%)0.01 (+95.59%)198.20 (+19.33%)160.00 (+4.92%)166.20 (+3.17%)106.50 (-14.66%)29.86 (+87.19%)
0a6c11c — 2025-12-04 00:39:100.04 (n/a)0.03 (n/a)0.03 (n/a)0.03 (n/a)0.00 (n/a)166.10 (n/a)152.50 (n/a)161.10 (n/a)124.80 (n/a)15.95 (n/a)

dequant_8_cols_2_channels_2048_tile_128_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.04 (+13.46%)0.04 (-0.92%)0.03 (-11.82%)0.03 (+14.75%)0.00 (-2.32%)168.20 (-12.89%)148.24 (+0.38%)154.40 (+13.45%)119.30 (-11.89%)18.57 (-26.89%)
897d04e — 2026-03-06 22:12:250.04 (n/a)0.04 (n/a)0.04 (n/a)0.03 (n/a)0.01 (n/a)193.10 (n/a)147.68 (n/a)136.10 (n/a)135.40 (n/a)25.41 (n/a)

dual_gemv_silu_mul_2048x2048_4tsi_512tso_4col0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:3838.08 (n/a)36.42 (n/a)36.81 (n/a)33.98 (n/a)1.52 (n/a)494.00 (n/a)461.52 (n/a)456.00 (n/a)440.80 (n/a)19.80 (n/a)

eltwise_add_1_cols_2_channels_2048_tile_2048

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.13 (-17.69%)0.11 (-14.81%)0.12 (-18.40%)0.08 (+0.61%)0.02 (-42.06%)144.90 (-0.55%)112.82 (+13.64%)102.30 (+22.66%)94.20 (+21.39%)20.43 (-28.81%)
897d04e — 2026-03-06 22:12:250.16 (n/a)0.13 (n/a)0.15 (n/a)0.08 (n/a)0.03 (n/a)145.70 (n/a)99.28 (n/a)83.40 (n/a)77.60 (n/a)28.70 (n/a)

eltwise_add_2_cols_2_channels_2048_tile_1024

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.13 (+6.50%)0.11 (+3.06%)0.11 (-2.20%)0.08 (+4.85%)0.02 (+0.14%)150.80 (-4.62%)118.50 (-3.28%)107.30 (+2.19%)95.20 (-6.11%)25.27 (-8.21%)
897d04e — 2026-03-06 22:12:250.12 (n/a)0.10 (n/a)0.12 (n/a)0.08 (n/a)0.02 (n/a)158.10 (n/a)122.52 (n/a)105.00 (n/a)101.40 (n/a)27.53 (n/a)

eltwise_add_4_cols_2_channels_2048_tile_512

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.12 (-25.68%)0.11 (-11.19%)0.11 (-19.62%)0.08 (+23.01%)0.02 (-57.53%)161.50 (-18.72%)118.46 (+2.35%)107.60 (+24.39%)106.70 (+34.55%)24.09 (-52.49%)
897d04e — 2026-03-06 22:12:250.15 (n/a)0.12 (n/a)0.14 (n/a)0.06 (n/a)0.04 (n/a)198.70 (n/a)115.74 (n/a)86.50 (n/a)79.30 (n/a)50.70 (n/a)

eltwise_add_8_cols_2_channels_2048_tile_256

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.14 (+27.93%)0.12 (+22.74%)0.13 (+22.20%)0.08 (+16.42%)0.02 (+37.85%)151.20 (-14.14%)102.94 (-17.74%)92.30 (-18.10%)87.10 (-21.81%)27.07 (-5.03%)
897d04e — 2026-03-06 22:12:250.11 (n/a)0.10 (n/a)0.11 (n/a)0.07 (n/a)0.02 (n/a)176.10 (n/a)125.14 (n/a)112.70 (n/a)111.40 (n/a)28.50 (n/a)

eltwise_mul_1_cols_2_channels_2048_tile_2048

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.12 (-18.30%)0.10 (-18.05%)0.12 (-14.77%)0.07 (+0.96%)0.02 (-35.47%)164.50 (-0.96%)121.06 (+17.90%)104.70 (+17.38%)102.50 (+22.46%)26.71 (-24.92%)
897d04e — 2026-03-06 22:12:250.15 (n/a)0.13 (n/a)0.14 (n/a)0.07 (n/a)0.03 (n/a)166.10 (n/a)102.68 (n/a)89.20 (n/a)83.70 (n/a)35.57 (n/a)

eltwise_mul_2_cols_2_channels_2048_tile_1024

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.12 (-18.56%)0.11 (-17.21%)0.12 (-18.87%)0.06 (-29.26%)0.03 (-2.58%)190.00 (+41.37%)119.06 (+23.92%)101.50 (+23.18%)100.00 (+22.70%)39.67 (+73.71%)
897d04e — 2026-03-06 22:12:250.15 (n/a)0.13 (n/a)0.15 (n/a)0.09 (n/a)0.03 (n/a)134.40 (n/a)96.08 (n/a)82.40 (n/a)81.50 (n/a)22.84 (n/a)

eltwise_mul_4_cols_2_channels_2048_tile_512

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.16 (+39.10%)0.15 (+44.73%)0.16 (+49.38%)0.10 (+62.49%)0.03 (+14.58%)120.90 (-38.44%)86.92 (-32.41%)77.30 (-33.02%)76.20 (-28.05%)19.21 (-49.89%)
897d04e — 2026-03-06 22:12:250.12 (n/a)0.10 (n/a)0.11 (n/a)0.06 (n/a)0.02 (n/a)196.40 (n/a)128.60 (n/a)115.40 (n/a)105.90 (n/a)38.33 (n/a)

eltwise_mul_8_cols_2_channels_2048_tile_256

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.13 (+22.52%)0.12 (+21.45%)0.13 (+22.75%)0.08 (+3.99%)0.02 (+75.73%)149.30 (-3.80%)103.78 (-15.86%)92.60 (-18.56%)91.60 (-18.36%)25.45 (+38.68%)
897d04e — 2026-03-06 22:12:250.11 (n/a)0.10 (n/a)0.11 (n/a)0.08 (n/a)0.01 (n/a)155.20 (n/a)123.34 (n/a)113.70 (n/a)112.20 (n/a)18.35 (n/a)

flowkv_decode_32h_8kv_64d_128s_32cs_4col0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.81 (n/a)0.78 (n/a)0.80 (n/a)0.70 (n/a)0.05 (n/a)388.90 (n/a)348.94 (n/a)339.90 (n/a)336.80 (n/a)22.38 (n/a)

fused_dequant_gemv_2048x2048_1tsi_512tso_4col_g32_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
116b247 — 2026-03-07 19:03:381.91 (n/a)1.88 (n/a)1.88 (n/a)1.84 (n/a)0.02 (n/a)1283.90 (n/a)1260.80 (n/a)1258.70 (n/a)1240.30 (n/a)16.10 (n/a)6.76 (n/a)6.65 (n/a)6.66 (n/a)6.53 (n/a)0.08 (n/a)

fused_qkv_proj_3072x2048_4tsi_768tso_4col0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
116b247 — 2026-03-07 19:03:3835.21 (n/a)33.92 (n/a)34.23 (n/a)31.91 (n/a)1.41 (n/a)394.60 (n/a)371.78 (n/a)367.90 (n/a)357.70 (n/a)15.70 (n/a)35.18 (n/a)33.89 (n/a)34.20 (n/a)31.88 (n/a)1.41 (n/a)

gelu_1_cols_1_channels_2048_tile_2048

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.10 (-1.79%)0.09 (-5.08%)0.10 (+2.83%)0.06 (-6.73%)0.02 (+22.47%)144.70 (+7.26%)102.14 (+7.65%)85.50 (-2.73%)79.00 (+1.80%)29.67 (+27.37%)
897d04e — 2026-03-06 22:12:250.11 (n/a)0.09 (n/a)0.09 (n/a)0.06 (n/a)0.02 (n/a)134.90 (n/a)94.88 (n/a)87.90 (n/a)77.60 (n/a)23.30 (n/a)

gelu_1_cols_2_channels_2048_tile_1024

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.11 (-5.86%)0.09 (-9.56%)0.09 (-10.87%)0.07 (-5.99%)0.02 (+7.12%)118.50 (+6.37%)94.12 (+11.25%)86.60 (+12.18%)77.20 (+6.19%)18.48 (+18.05%)
897d04e — 2026-03-06 22:12:250.11 (n/a)0.10 (n/a)0.11 (n/a)0.07 (n/a)0.02 (n/a)111.40 (n/a)84.60 (n/a)77.20 (n/a)72.70 (n/a)15.66 (n/a)

gelu_2_cols_1_channels_2048_tile_1024

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.11 (+2.15%)0.09 (-3.98%)0.11 (+2.43%)0.05 (-2.80%)0.03 (+12.84%)160.40 (+2.89%)100.46 (+5.64%)77.50 (-2.39%)75.10 (-2.09%)36.88 (+7.90%)
897d04e — 2026-03-06 22:12:250.11 (n/a)0.09 (n/a)0.10 (n/a)0.05 (n/a)0.02 (n/a)155.90 (n/a)95.10 (n/a)79.40 (n/a)76.70 (n/a)34.18 (n/a)

gelu_2_cols_2_channels_2048_tile_512

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.11 (-3.45%)0.10 (+0.76%)0.11 (-1.79%)0.06 (+18.71%)0.02 (-23.70%)127.80 (-15.75%)88.72 (-4.23%)77.90 (+1.70%)77.60 (+3.60%)21.94 (-33.99%)
897d04e — 2026-03-06 22:12:250.11 (n/a)0.10 (n/a)0.11 (n/a)0.05 (n/a)0.02 (n/a)151.70 (n/a)92.64 (n/a)76.60 (n/a)74.90 (n/a)33.24 (n/a)

gelu_4_cols_1_channels_2048_tile_512

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.10 (-5.01%)0.09 (-6.39%)0.09 (-8.41%)0.05 (-2.65%)0.02 (-12.49%)155.40 (+2.71%)100.68 (+5.80%)91.50 (+9.19%)78.60 (+5.22%)31.05 (-2.85%)
897d04e — 2026-03-06 22:12:250.11 (n/a)0.09 (n/a)0.10 (n/a)0.05 (n/a)0.02 (n/a)151.30 (n/a)95.16 (n/a)83.80 (n/a)74.70 (n/a)31.97 (n/a)

gelu_4_cols_2_channels_2048_tile_256

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.10 (+3.13%)0.08 (-4.93%)0.09 (-4.94%)0.04 (-12.50%)0.03 (+9.82%)207.40 (+14.27%)113.10 (+8.33%)89.30 (+5.18%)80.40 (-3.02%)53.52 (+24.00%)
897d04e — 2026-03-06 22:12:250.10 (n/a)0.09 (n/a)0.10 (n/a)0.05 (n/a)0.02 (n/a)181.50 (n/a)104.40 (n/a)84.90 (n/a)82.90 (n/a)43.16 (n/a)

gelu_8_cols_1_channels_2048_tile_256

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.10 (-1.80%)0.08 (-2.81%)0.09 (-0.11%)0.06 (+1.97%)0.02 (+4.99%)140.50 (-1.95%)102.84 (+3.07%)91.40 (+0.11%)83.10 (+1.84%)24.73 (-0.91%)
897d04e — 2026-03-06 22:12:250.10 (n/a)0.09 (n/a)0.09 (n/a)0.06 (n/a)0.02 (n/a)143.30 (n/a)99.78 (n/a)91.30 (n/a)81.60 (n/a)24.96 (n/a)

gelu_8_cols_2_channels_2048_tile_128

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.08 (-1.98%)0.07 (-0.54%)0.08 (+5.46%)0.05 (-1.04%)0.02 (+0.89%)175.50 (+1.04%)117.66 (+0.67%)100.70 (-5.18%)100.20 (+2.04%)32.69 (+2.08%)
897d04e — 2026-03-06 22:12:250.08 (n/a)0.07 (n/a)0.08 (n/a)0.05 (n/a)0.01 (n/a)173.70 (n/a)116.88 (n/a)106.20 (n/a)98.20 (n/a)32.02 (n/a)

gemm_1792x896x1152_64x32x48_8cols_ccolmaj

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
116b247 — 2026-03-07 19:03:385.16 (+3.35%)4.92 (+0.77%)4.99 (+0.41%)4.47 (-0.86%)0.26 (+25.84%)2101.70 (+0.87%)1915.48 (-0.68%)1885.90 (-0.41%)1821.30 (-3.24%)107.69 (+24.11%)2031.14 (+3.35%)1935.92 (+0.77%)1961.64 (+0.41%)1760.14 (-0.86%)102.65 (+25.84%)
897d04e — 2026-03-06 22:12:255.00 (n/a)4.88 (n/a)4.97 (n/a)4.51 (n/a)0.21 (n/a)2083.60 (n/a)1928.60 (n/a)1893.70 (n/a)1882.30 (n/a)86.77 (n/a)1965.38 (n/a)1921.12 (n/a)1953.55 (n/a)1775.47 (n/a)81.57 (n/a)

gemm_192x384x64_48x96x16_4cols

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
116b247 — 2026-03-07 19:03:382.01 (-1.82%)1.68 (+1.35%)1.87 (+0.09%)1.08 (+36.94%)0.38 (-28.26%)205.60 (-26.99%)138.56 (-8.67%)118.50 (-0.08%)109.90 (+1.85%)39.40 (-46.69%)85.83 (-1.82%)71.73 (+1.35%)79.63 (+0.09%)45.90 (+36.94%)16.13 (-28.26%)
897d04e — 2026-03-06 22:12:252.05 (n/a)1.66 (n/a)1.86 (n/a)0.79 (n/a)0.53 (n/a)281.60 (n/a)151.72 (n/a)118.60 (n/a)107.90 (n/a)73.91 (n/a)87.43 (n/a)70.78 (n/a)79.55 (n/a)33.52 (n/a)22.49 (n/a)

gemm_192x384x64_48x96x16_4cols_bcolmaj_ccolmaj

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
116b247 — 2026-03-07 19:03:381.97 (-11.74%)1.70 (-6.60%)1.89 (+0.89%)0.97 (-26.70%)0.42 (+28.92%)229.00 (+36.47%)139.70 (+11.92%)117.30 (-0.85%)112.30 (+13.32%)50.24 (+96.43%)84.06 (-11.74%)72.72 (-6.60%)80.48 (+0.89%)41.22 (-26.70%)18.00 (+28.92%)
897d04e — 2026-03-06 22:12:252.23 (n/a)1.82 (n/a)1.87 (n/a)1.32 (n/a)0.33 (n/a)167.80 (n/a)124.82 (n/a)118.30 (n/a)99.10 (n/a)25.58 (n/a)95.25 (n/a)77.86 (n/a)79.77 (n/a)56.23 (n/a)13.96 (n/a)

gemm_2048x2048x2048_64x64x32_8_cols_0_bcolmaj_0_ccolmaj_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)GFLOP/s (max)GFLOP/s (mean)GFLOP/s (median)GFLOP/s (min)GFLOP/s (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
3a053d3 — 2025-11-26 01:12:163.72 (-0.11%)3.55 (-2.75%)3.56 (-2.75%)3.36 (-5.33%)0.12 (+76.10%)2725.23 (-0.11%)2602.52 (-2.75%)2612.11 (-2.75%)2459.89 (-5.33%)86.27 (+76.10%)6984.00 (+5.63%)6608.60 (+2.91%)6577.00 (+2.83%)6304.00 (+0.11%)221.87 (+87.15%)
9e98049 — 2025-11-25 20:50:553.72 (n/a)3.65 (n/a)3.66 (n/a)3.54 (n/a)0.07 (n/a)2728.26 (n/a)2676.06 (n/a)2686.03 (n/a)2598.29 (n/a)48.99 (n/a)6612.00 (n/a)6422.00 (n/a)6396.00 (n/a)6297.00 (n/a)118.55 (n/a)

gemm_2048x2048x2048_64x64x32_8_cols_0_bcolmaj_1_ccolmaj_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)GFLOP/s (max)GFLOP/s (mean)GFLOP/s (median)GFLOP/s (min)GFLOP/s (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
3a053d3 — 2025-11-26 01:12:163.40 (-3.14%)3.38 (-0.82%)3.38 (-0.82%)3.35 (+4.93%)0.02 (-84.51%)2493.81 (-3.14%)2475.27 (-0.82%)2479.06 (-0.82%)2453.92 (+4.93%)13.33 (-84.51%)7001.00 (-4.70%)6940.80 (+0.71%)6930.00 (+0.83%)6889.00 (+3.24%)37.43 (-84.76%)
9e98049 — 2025-11-25 20:50:553.51 (n/a)3.40 (n/a)3.41 (n/a)3.19 (n/a)0.12 (n/a)2574.53 (n/a)2495.72 (n/a)2499.62 (n/a)2338.67 (n/a)86.02 (n/a)7346.00 (n/a)6892.20 (n/a)6873.00 (n/a)6673.00 (n/a)245.57 (n/a)

gemm_2048x2048x2048_64x64x32_8_cols_1_bcolmaj_0_ccolmaj_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)GFLOP/s (max)GFLOP/s (mean)GFLOP/s (median)GFLOP/s (min)GFLOP/s (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
3a053d3 — 2025-11-26 01:12:163.02 (+0.52%)2.87 (-2.05%)2.86 (-2.24%)2.70 (-6.16%)0.11 (+131.66%)2216.19 (+0.52%)2104.44 (-2.05%)2094.08 (-2.24%)1980.84 (-6.16%)80.25 (+131.67%)8673.00 (+6.56%)8175.60 (+2.21%)8204.00 (+2.29%)7752.00 (-0.51%)314.03 (+145.36%)
9e98049 — 2025-11-25 20:50:553.01 (n/a)2.93 (n/a)2.92 (n/a)2.88 (n/a)0.05 (n/a)2204.81 (n/a)2148.41 (n/a)2142.13 (n/a)2110.81 (n/a)34.64 (n/a)8139.00 (n/a)7998.60 (n/a)8020.00 (n/a)7792.00 (n/a)127.99 (n/a)

gemm_2048x2048x2048_64x64x64_1cols

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
116b247 — 2026-03-07 19:03:380.52 (-0.77%)0.52 (-0.22%)0.52 (-0.04%)0.52 (-0.19%)0.00 (-75.37%)48495.20 (+0.19%)48422.36 (+0.21%)48410.30 (+0.04%)48384.00 (+0.77%)42.48 (-75.13%)355.07 (-0.77%)354.79 (-0.22%)354.88 (-0.04%)354.26 (-0.19%)0.31 (-75.37%)
897d04e — 2026-03-06 22:12:250.52 (n/a)0.52 (n/a)0.52 (n/a)0.52 (n/a)0.00 (n/a)48401.80 (n/a)48318.62 (n/a)48391.10 (n/a)48013.20 (n/a)170.80 (n/a)357.82 (n/a)355.56 (n/a)355.02 (n/a)354.94 (n/a)1.26 (n/a)

gemm_2048x2048x2048_64x64x64_2_cols_0_bcolmaj_0_ccolmaj_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
130b6ea — 2025-12-05 21:15:491.04 (+0.94%)1.03 (+0.23%)1.03 (+0.09%)1.02 (+0.07%)0.00 (+197.51%)24576.00 (-0.07%)24485.82 (-0.23%)24530.40 (-0.09%)24266.40 (-0.94%)111.26 (+194.35%)
0a6c11c — 2025-12-04 00:39:101.03 (n/a)1.03 (n/a)1.03 (n/a)1.02 (n/a)0.00 (n/a)24594.30 (n/a)24542.14 (n/a)24551.90 (n/a)24495.50 (n/a)37.80 (n/a)

gemm_2048x2048x2048_64x64x64_2_cols_0_bcolmaj_0_ccolmaj_0_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
a4b6ffe — 2026-01-05 19:24:051.04 (+0.28%)1.03 (-0.15%)1.03 (+0.05%)1.02 (-0.52%)0.01 (+42.72%)24719.80 (+0.52%)24512.42 (+0.16%)24529.20 (-0.05%)24092.60 (-0.28%)252.92 (+43.15%)713.08 (+0.28%)700.92 (-0.15%)700.38 (+0.05%)694.98 (-0.52%)7.30 (+42.71%)
50ad37c — 2025-12-29 20:35:531.04 (n/a)1.03 (n/a)1.03 (n/a)1.02 (n/a)0.01 (n/a)24592.30 (n/a)24473.70 (n/a)24540.90 (n/a)24160.80 (n/a)176.68 (n/a)711.06 (n/a)702.00 (n/a)700.05 (n/a)698.59 (n/a)5.12 (n/a)

gemm_2048x2048x2048_64x64x64_2_cols_0_bcolmaj_1_ccolmaj_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
130b6ea — 2025-12-05 21:15:490.99 (+1.69%)0.97 (+0.29%)0.97 (-0.04%)0.97 (-0.15%)0.01 (+434.92%)26025.30 (+0.15%)25839.34 (-0.29%)25916.70 (+0.04%)25446.00 (-1.67%)202.05 (+425.44%)
0a6c11c — 2025-12-04 00:39:100.97 (n/a)0.97 (n/a)0.97 (n/a)0.97 (n/a)0.00 (n/a)25986.90 (n/a)25913.52 (n/a)25905.70 (n/a)25877.00 (n/a)38.45 (n/a)

gemm_2048x2048x2048_64x64x64_2_cols_0_bcolmaj_1_ccolmaj_0_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
a4b6ffe — 2026-01-05 19:24:050.97 (-0.01%)0.97 (+0.03%)0.97 (+0.04%)0.97 (+0.00%)0.00 (-6.43%)25927.90 (-0.00%)25903.20 (-0.03%)25898.60 (-0.04%)25892.80 (+0.01%)14.05 (-6.56%)663.50 (-0.01%)663.23 (+0.03%)663.35 (+0.04%)662.60 (+0.00%)0.36 (-6.43%)
50ad37c — 2025-12-29 20:35:530.97 (n/a)0.97 (n/a)0.97 (n/a)0.97 (n/a)0.00 (n/a)25928.10 (n/a)25910.14 (n/a)25908.10 (n/a)25891.20 (n/a)15.04 (n/a)663.54 (n/a)663.06 (n/a)663.11 (n/a)662.60 (n/a)0.38 (n/a)

gemm_2048x2048x2048_64x64x64_2_cols_1_bcolmaj_0_ccolmaj_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
130b6ea — 2025-12-05 21:15:490.92 (-1.02%)0.92 (-0.66%)0.92 (-0.10%)0.91 (-0.78%)0.01 (-4.54%)27751.40 (+0.79%)27480.74 (+0.66%)27354.80 (+0.10%)27320.50 (+1.03%)182.69 (-2.84%)
0a6c11c — 2025-12-04 00:39:100.93 (n/a)0.92 (n/a)0.92 (n/a)0.91 (n/a)0.01 (n/a)27534.80 (n/a)27299.32 (n/a)27326.70 (n/a)27042.30 (n/a)188.04 (n/a)

gemm_2048x2048x2048_64x64x64_2_cols_1_bcolmaj_0_ccolmaj_0_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
a4b6ffe — 2026-01-05 19:24:050.92 (+0.11%)0.91 (-0.12%)0.91 (-0.21%)0.91 (-0.24%)0.00 (+30.34%)27723.50 (+0.24%)27548.94 (+0.12%)27522.90 (+0.21%)27403.60 (-0.11%)124.26 (+30.58%)626.92 (+0.11%)623.62 (-0.12%)624.20 (-0.21%)619.69 (-0.24%)2.81 (+30.34%)
50ad37c — 2025-12-29 20:35:530.92 (n/a)0.91 (n/a)0.92 (n/a)0.91 (n/a)0.00 (n/a)27657.10 (n/a)27515.58 (n/a)27464.70 (n/a)27434.60 (n/a)95.16 (n/a)626.21 (n/a)624.37 (n/a)625.52 (n/a)621.17 (n/a)2.16 (n/a)

gemm_2048x2048x2048_64x64x64_2cols_bcolmaj

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
116b247 — 2026-03-07 19:03:380.92 (+0.64%)0.91 (+0.83%)0.91 (+0.14%)0.90 (+2.15%)0.01 (-44.80%)27848.00 (-2.11%)27666.40 (-0.83%)27734.80 (-0.14%)27344.70 (-0.64%)192.63 (-46.44%)628.27 (+0.64%)620.99 (+0.83%)619.43 (+0.14%)616.92 (+2.15%)4.35 (-44.80%)
897d04e — 2026-03-06 22:12:250.91 (n/a)0.90 (n/a)0.91 (n/a)0.88 (n/a)0.01 (n/a)28447.10 (n/a)27899.18 (n/a)27774.40 (n/a)27519.50 (n/a)359.64 (n/a)624.28 (n/a)615.87 (n/a)618.55 (n/a)603.92 (n/a)7.88 (n/a)

gemm_2048x2048x2048_64x64x64_8_cols_0_bcolmaj_0_ccolmaj_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)GFLOP/s (max)GFLOP/s (mean)GFLOP/s (median)GFLOP/s (min)GFLOP/s (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
31ab24e — 2025-11-22 02:01:563.65 (-1.06%)3.59 (+0.83%)3.63 (-0.26%)3.45 (+2.66%)0.07 (-44.71%)2678.08 (-1.06%)2632.13 (+0.83%)2658.60 (-0.26%)2528.31 (+2.66%)53.86 (-44.71%)6795.00 (-2.59%)6529.80 (-0.92%)6462.00 (+0.26%)6415.00 (+1.07%)137.14 (-45.27%)
d9a64c5 — 2025-11-21 20:49:353.69 (n/a)3.56 (n/a)3.64 (n/a)3.36 (n/a)0.13 (n/a)2706.77 (n/a)2610.50 (n/a)2665.61 (n/a)2462.71 (n/a)97.41 (n/a)6976.00 (n/a)6590.40 (n/a)6445.00 (n/a)6347.00 (n/a)250.58 (n/a)

gemm_2048x2048x2048_64x64x64_8_cols_0_bcolmaj_0_ccolmaj_0_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
d48746f — 2026-01-12 22:17:103.98 (+0.20%)3.84 (-2.95%)3.76 (-5.27%)3.72 (-4.97%)0.13 (+473.97%)6760.00 (+5.23%)6559.50 (+3.13%)6697.10 (+5.56%)6321.00 (-0.20%)218.68 (+499.40%)2717.91 (+0.20%)2621.44 (-2.95%)2565.27 (-5.27%)2541.40 (-4.97%)88.37 (+473.97%)
331dcca — 2026-01-08 18:34:103.97 (n/a)3.96 (n/a)3.97 (n/a)3.92 (n/a)0.02 (n/a)6423.80 (n/a)6360.32 (n/a)6344.30 (n/a)6333.90 (n/a)36.48 (n/a)2712.39 (n/a)2701.17 (n/a)2707.91 (n/a)2674.40 (n/a)15.40 (n/a)

gemm_2048x2048x2048_64x64x64_8_cols_0_bcolmaj_1_ccolmaj_0_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
d48746f — 2026-01-12 22:17:103.75 (+0.28%)3.70 (+1.25%)3.71 (-0.21%)3.65 (+4.77%)0.04 (-66.90%)6901.80 (-4.56%)6795.46 (-1.30%)6790.20 (+0.21%)6715.90 (-0.28%)67.05 (-68.40%)2558.08 (+0.28%)2528.33 (+1.25%)2530.10 (-0.21%)2489.18 (+4.77%)24.81 (-66.90%)
331dcca — 2026-01-08 18:34:103.74 (n/a)3.66 (n/a)3.71 (n/a)3.48 (n/a)0.11 (n/a)7231.20 (n/a)6885.18 (n/a)6775.70 (n/a)6734.90 (n/a)212.21 (n/a)2550.87 (n/a)2497.04 (n/a)2535.50 (n/a)2375.78 (n/a)74.96 (n/a)

gemm_2048x2048x2048_64x64x64_8_cols_1_bcolmaj_0_ccolmaj_0_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
d48746f — 2026-01-12 22:17:103.62 (-0.77%)3.54 (+0.14%)3.60 (-0.04%)3.44 (+2.17%)0.09 (-25.77%)7307.40 (-2.13%)7103.46 (-0.18%)6982.50 (+0.04%)6948.20 (+0.77%)182.95 (-26.55%)2472.55 (-0.77%)2419.80 (+0.14%)2460.41 (-0.04%)2351.04 (+2.17%)61.76 (-25.77%)
331dcca — 2026-01-08 18:34:103.65 (n/a)3.54 (n/a)3.61 (n/a)3.37 (n/a)0.12 (n/a)7466.20 (n/a)7116.40 (n/a)6979.60 (n/a)6895.10 (n/a)249.09 (n/a)2491.61 (n/a)2416.45 (n/a)2461.45 (n/a)2301.00 (n/a)83.20 (n/a)

gemm_2048x2048x2048_64x64x64_8cols_bcolmaj_ccolmaj

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
116b247 — 2026-03-07 19:03:383.45 (-1.74%)3.31 (-5.24%)3.28 (-5.76%)3.23 (-7.27%)0.09 (+658.86%)7803.00 (+7.84%)7618.82 (+5.60%)7668.60 (+6.11%)7298.80 (+1.77%)212.99 (+734.82%)2353.81 (-1.74%)2256.36 (-5.24%)2240.28 (-5.76%)2201.71 (-7.27%)64.10 (+658.85%)
897d04e — 2026-03-06 22:12:253.51 (n/a)3.49 (n/a)3.48 (n/a)3.48 (n/a)0.01 (n/a)7235.40 (n/a)7215.10 (n/a)7227.20 (n/a)7171.80 (n/a)25.51 (n/a)2395.46 (n/a)2381.12 (n/a)2377.13 (n/a)2374.41 (n/a)8.45 (n/a)

gemm_384x1536x1792_32x48x64_4cols_bcolmaj

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
116b247 — 2026-03-07 19:03:384.34 (+0.04%)4.20 (+0.70%)4.28 (-0.79%)3.85 (+9.22%)0.20 (-43.56%)2091.10 (-8.45%)1925.26 (-1.16%)1881.40 (+0.80%)1855.40 (-0.04%)97.15 (-48.33%)1139.32 (+0.04%)1100.12 (+0.70%)1123.60 (-0.79%)1010.91 (+9.22%)52.71 (-43.56%)
897d04e — 2026-03-06 22:12:254.34 (n/a)4.17 (n/a)4.32 (n/a)3.53 (n/a)0.36 (n/a)2284.00 (n/a)1947.78 (n/a)1866.50 (n/a)1856.20 (n/a)188.02 (n/a)1138.86 (n/a)1092.52 (n/a)1132.55 (n/a)925.53 (n/a)93.39 (n/a)

gemm_896x1792x640_32x64x80_8cols_ccolmaj

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
116b247 — 2026-03-07 19:03:387.10 (+5.62%)6.53 (+3.45%)6.82 (+3.15%)5.23 (+0.62%)0.75 (+16.42%)1273.00 (-0.61%)1031.94 (-3.07%)975.10 (-3.05%)936.80 (-5.32%)137.58 (+11.09%)2193.95 (+5.62%)2016.37 (+3.45%)2107.79 (+3.15%)1614.47 (+0.62%)232.52 (+16.42%)
897d04e — 2026-03-06 22:12:256.72 (n/a)6.31 (n/a)6.61 (n/a)5.19 (n/a)0.65 (n/a)1280.80 (n/a)1064.58 (n/a)1005.80 (n/a)989.40 (n/a)123.84 (n/a)2077.17 (n/a)1949.09 (n/a)2043.44 (n/a)1604.60 (n/a)199.72 (n/a)

layer_norm_1_cols_1_channels_2048_tile_2048

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.10 (-3.22%)0.08 (-5.73%)0.09 (+5.59%)0.04 (-24.74%)0.03 (+12.31%)198.80 (+32.80%)117.06 (+11.15%)89.60 (-5.39%)81.00 (+3.32%)49.36 (+58.69%)
897d04e — 2026-03-06 22:12:250.10 (n/a)0.08 (n/a)0.09 (n/a)0.05 (n/a)0.02 (n/a)149.70 (n/a)105.32 (n/a)94.70 (n/a)78.40 (n/a)31.11 (n/a)

layer_norm_1_cols_2_channels_2048_tile_1024

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.10 (-4.41%)0.09 (+0.42%)0.10 (-1.70%)0.05 (-11.07%)0.02 (-0.44%)157.10 (+12.46%)95.74 (+0.84%)79.10 (+1.67%)78.30 (+4.54%)34.44 (+22.03%)
897d04e — 2026-03-06 22:12:250.11 (n/a)0.09 (n/a)0.11 (n/a)0.06 (n/a)0.02 (n/a)139.70 (n/a)94.94 (n/a)77.80 (n/a)74.90 (n/a)28.23 (n/a)

layer_norm_2_cols_1_channels_2048_tile_1024

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.12 (+9.54%)0.10 (+10.15%)0.10 (+8.24%)0.06 (+3.47%)0.02 (+18.12%)131.70 (-3.30%)87.44 (-8.29%)78.90 (-7.61%)71.10 (-8.61%)24.98 (+5.70%)
897d04e — 2026-03-06 22:12:250.11 (n/a)0.09 (n/a)0.10 (n/a)0.06 (n/a)0.02 (n/a)136.20 (n/a)95.34 (n/a)85.40 (n/a)77.80 (n/a)23.63 (n/a)

layer_norm_2_cols_2_channels_2048_tile_512

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.11 (+0.20%)0.10 (+17.63%)0.11 (+21.61%)0.06 (+59.86%)0.02 (-23.39%)134.50 (-37.44%)88.54 (-21.99%)77.70 (-17.78%)75.90 (-0.13%)25.71 (-55.16%)
897d04e — 2026-03-06 22:12:250.11 (n/a)0.08 (n/a)0.09 (n/a)0.04 (n/a)0.03 (n/a)215.00 (n/a)113.50 (n/a)94.50 (n/a)76.00 (n/a)57.35 (n/a)

layer_norm_4_cols_1_channels_2048_tile_512

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.13 (+19.85%)0.10 (+14.65%)0.10 (+14.45%)0.06 (+29.03%)0.03 (+7.09%)146.90 (-22.48%)90.72 (-15.06%)79.20 (-12.68%)65.10 (-16.54%)32.28 (-30.62%)
897d04e — 2026-03-06 22:12:250.11 (n/a)0.09 (n/a)0.09 (n/a)0.04 (n/a)0.02 (n/a)189.50 (n/a)106.80 (n/a)90.70 (n/a)78.00 (n/a)46.53 (n/a)

layer_norm_4_cols_2_channels_2048_tile_256

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.10 (+1.43%)0.07 (-7.21%)0.08 (-7.12%)0.04 (-19.78%)0.03 (+29.76%)223.60 (+24.64%)125.50 (+14.44%)101.00 (+7.68%)84.20 (-1.41%)58.45 (+48.92%)
897d04e — 2026-03-06 22:12:250.10 (n/a)0.08 (n/a)0.09 (n/a)0.05 (n/a)0.02 (n/a)179.40 (n/a)109.66 (n/a)93.80 (n/a)85.40 (n/a)39.25 (n/a)

layer_norm_8_cols_1_channels_2048_tile_256

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.10 (+8.20%)0.08 (+10.95%)0.10 (+19.41%)0.04 (-16.88%)0.02 (+49.92%)189.20 (+20.28%)108.34 (-4.14%)84.90 (-16.27%)78.40 (-7.66%)46.46 (+66.94%)
897d04e — 2026-03-06 22:12:250.10 (n/a)0.08 (n/a)0.08 (n/a)0.05 (n/a)0.02 (n/a)157.30 (n/a)113.02 (n/a)101.40 (n/a)84.90 (n/a)27.83 (n/a)

layer_norm_8_cols_2_channels_2048_tile_128

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.08 (-2.38%)0.07 (-7.69%)0.07 (-19.93%)0.04 (-2.77%)0.02 (-5.21%)203.60 (+2.83%)130.96 (+7.73%)123.30 (+24.80%)99.00 (+2.48%)42.90 (-1.39%)
897d04e — 2026-03-06 22:12:250.08 (n/a)0.07 (n/a)0.08 (n/a)0.04 (n/a)0.02 (n/a)198.00 (n/a)121.56 (n/a)98.80 (n/a)96.60 (n/a)43.50 (n/a)

matrix_vector_mul_128x128_32_1col

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)
130b6ea — 2025-12-05 21:15:490.37 (+17.13%)0.30 (+24.87%)0.30 (+38.03%)0.23 (+12.47%)0.06 (+35.23%)
0a6c11c — 2025-12-04 00:39:100.32 (n/a)0.24 (n/a)0.22 (n/a)0.20 (n/a)0.04 (n/a)

matrix_vector_mul_128x128_32_1col0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
331dcca — 2026-01-08 18:34:100.26 (-1.81%)0.24 (+4.06%)0.24 (+0.52%)0.21 (+16.97%)0.02 (-48.16%)0.26 (-1.81%)0.24 (+4.06%)0.24 (+0.52%)0.20 (+16.97%)0.02 (-48.16%)
a4b6ffe — 2026-01-05 19:24:050.26 (n/a)0.23 (n/a)0.24 (n/a)0.18 (n/a)0.04 (n/a)0.26 (n/a)0.23 (n/a)0.24 (n/a)0.17 (n/a)0.04 (n/a)

matrix_vector_mul_128x128_32tsi_128tso_1col0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
116b247 — 2026-03-07 19:03:380.35 (-5.43%)0.31 (+0.10%)0.34 (-1.42%)0.25 (+15.14%)0.04 (-30.16%)0.35 (-5.43%)0.31 (+0.10%)0.33 (-1.42%)0.25 (+15.14%)0.04 (-30.16%)
897d04e — 2026-03-06 22:12:250.37 (n/a)0.31 (n/a)0.34 (n/a)0.22 (n/a)0.06 (n/a)0.36 (n/a)0.31 (n/a)0.34 (n/a)0.21 (n/a)0.06 (n/a)

matrix_vector_mul_2048x8192_1_1col

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)
130b6ea — 2025-12-05 21:15:4913.57 (+3.47%)13.15 (+1.09%)13.10 (+0.16%)12.52 (-1.58%)0.37 (+153.19%)
0a6c11c — 2025-12-04 00:39:1013.11 (n/a)13.01 (n/a)13.07 (n/a)12.72 (n/a)0.15 (n/a)

matrix_vector_mul_2048x8192_1_1col0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
331dcca — 2026-01-08 18:34:1013.43 (-0.07%)13.30 (+4.62%)13.27 (+1.29%)13.16 (+16.70%)0.12 (-86.69%)13.42 (-0.07%)13.29 (+4.62%)13.26 (+1.29%)13.16 (+16.70%)0.12 (-86.69%)
a4b6ffe — 2026-01-05 19:24:0513.44 (n/a)12.71 (n/a)13.10 (n/a)11.28 (n/a)0.87 (n/a)13.43 (n/a)12.70 (n/a)13.09 (n/a)11.27 (n/a)0.87 (n/a)

matrix_vector_mul_2048x8192_1_2col

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)
130b6ea — 2025-12-05 21:15:4925.06 (+2.47%)23.71 (+1.81%)24.13 (+0.50%)20.90 (+5.41%)1.48 (-15.60%)
0a6c11c — 2025-12-04 00:39:1024.46 (n/a)23.29 (n/a)24.01 (n/a)19.83 (n/a)1.75 (n/a)

matrix_vector_mul_2048x8192_1_2col0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
331dcca — 2026-01-08 18:34:1024.46 (-0.60%)21.07 (-8.03%)19.75 (-17.83%)18.08 (+1.18%)3.03 (+7.07%)24.44 (-0.60%)21.06 (-8.03%)19.74 (-17.83%)18.07 (+1.18%)3.03 (+7.07%)
a4b6ffe — 2026-01-05 19:24:0524.60 (n/a)22.91 (n/a)24.04 (n/a)17.87 (n/a)2.83 (n/a)24.59 (n/a)22.90 (n/a)24.02 (n/a)17.86 (n/a)2.83 (n/a)

matrix_vector_mul_2048x8192_1_4col

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)
130b6ea — 2025-12-05 21:15:4942.05 (+1.54%)40.50 (+12.60%)40.49 (+2.36%)38.98 (+97.53%)0.97 (-88.09%)
0a6c11c — 2025-12-04 00:39:1041.41 (n/a)35.97 (n/a)39.55 (n/a)19.73 (n/a)8.15 (n/a)

matrix_vector_mul_2048x8192_1_4col0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
331dcca — 2026-01-08 18:34:1042.58 (+1.51%)41.37 (+14.29%)41.25 (+4.89%)40.33 (+91.42%)0.92 (-89.18%)42.56 (+1.51%)41.34 (+14.29%)41.23 (+4.89%)40.30 (+91.42%)0.92 (-89.18%)
a4b6ffe — 2026-01-05 19:24:0541.95 (n/a)36.20 (n/a)39.33 (n/a)21.07 (n/a)8.54 (n/a)41.92 (n/a)36.17 (n/a)39.31 (n/a)21.05 (n/a)8.53 (n/a)

matrix_vector_mul_2048x8192_1_8col

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)
130b6ea — 2025-12-05 21:15:4946.62 (+3.11%)44.82 (+0.17%)44.78 (+0.17%)42.60 (-3.80%)1.60 (+367.72%)
0a6c11c — 2025-12-04 00:39:1045.21 (n/a)44.75 (n/a)44.70 (n/a)44.29 (n/a)0.34 (n/a)

matrix_vector_mul_2048x8192_1_8col0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
331dcca — 2026-01-08 18:34:1046.04 (+4.14%)45.06 (+3.47%)44.55 (+2.76%)44.30 (+2.84%)0.87 (+66.58%)46.01 (+4.14%)45.03 (+3.47%)44.52 (+2.76%)44.27 (+2.84%)0.87 (+66.58%)
a4b6ffe — 2026-01-05 19:24:0544.21 (n/a)43.55 (n/a)43.35 (n/a)43.07 (n/a)0.52 (n/a)44.18 (n/a)43.52 (n/a)43.32 (n/a)43.05 (n/a)0.52 (n/a)

matrix_vector_mul_2048x8192_1tsi_1024tso_2col0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
116b247 — 2026-03-07 19:03:3824.35 (-0.51%)23.14 (-4.24%)24.07 (-0.52%)19.27 (-19.17%)2.17 (+779.96%)24.33 (-0.51%)23.12 (-4.24%)24.06 (-0.52%)19.26 (-19.17%)2.17 (+779.96%)
897d04e — 2026-03-06 22:12:2524.47 (n/a)24.16 (n/a)24.20 (n/a)23.84 (n/a)0.25 (n/a)24.46 (n/a)24.15 (n/a)24.18 (n/a)23.83 (n/a)0.25 (n/a)

matrix_vector_mul_2048x8192_1tsi_2048tso_1col0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
116b247 — 2026-03-07 19:03:3813.45 (+1.52%)13.09 (+0.52%)13.11 (+0.06%)12.67 (+0.86%)0.28 (+3.59%)13.44 (+1.52%)13.08 (+0.52%)13.10 (+0.06%)12.66 (+0.86%)0.28 (+3.59%)
897d04e — 2026-03-06 22:12:2513.25 (n/a)13.02 (n/a)13.10 (n/a)12.56 (n/a)0.27 (n/a)13.24 (n/a)13.01 (n/a)13.09 (n/a)12.55 (n/a)0.27 (n/a)

matrix_vector_mul_2048x8192_1tsi_256tso_8col0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
116b247 — 2026-03-07 19:03:3845.40 (-4.84%)44.80 (-1.29%)44.83 (-0.85%)44.37 (+2.07%)0.43 (-71.97%)45.37 (-4.84%)44.77 (-1.29%)44.81 (-0.85%)44.35 (+2.07%)0.43 (-71.97%)
897d04e — 2026-03-06 22:12:2547.71 (n/a)45.38 (n/a)45.22 (n/a)43.47 (n/a)1.52 (n/a)47.68 (n/a)45.35 (n/a)45.19 (n/a)43.45 (n/a)1.52 (n/a)

matrix_vector_mul_2048x8192_1tsi_512tso_4col0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
116b247 — 2026-03-07 19:03:3841.92 (+2.65%)39.87 (+0.88%)39.74 (+0.86%)38.20 (-1.72%)1.56 (+102.52%)41.89 (+2.65%)39.85 (+0.88%)39.71 (+0.86%)38.17 (-1.72%)1.56 (+102.52%)
897d04e — 2026-03-06 22:12:2540.84 (n/a)39.53 (n/a)39.40 (n/a)38.87 (n/a)0.77 (n/a)40.81 (n/a)39.50 (n/a)39.37 (n/a)38.84 (n/a)0.77 (n/a)

matrix_vector_mul_8192x2048_4_1col

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)
130b6ea — 2025-12-05 21:15:4913.29 (+1.63%)13.19 (+6.08%)13.19 (+1.17%)13.09 (+26.86%)0.08 (-92.94%)
0a6c11c — 2025-12-04 00:39:1013.08 (n/a)12.44 (n/a)13.03 (n/a)10.32 (n/a)1.07 (n/a)

matrix_vector_mul_8192x2048_4_1col0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
331dcca — 2026-01-08 18:34:1013.68 (+3.42%)13.30 (+8.54%)13.38 (+7.25%)12.72 (+14.96%)0.35 (-66.09%)13.67 (+3.42%)13.29 (+8.54%)13.38 (+7.25%)12.71 (+14.96%)0.35 (-66.09%)
a4b6ffe — 2026-01-05 19:24:0513.23 (n/a)12.25 (n/a)12.48 (n/a)11.07 (n/a)1.04 (n/a)13.22 (n/a)12.24 (n/a)12.47 (n/a)11.06 (n/a)1.04 (n/a)

matrix_vector_mul_8192x2048_4_2col

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)
130b6ea — 2025-12-05 21:15:4925.58 (+4.24%)24.23 (+6.59%)24.03 (+0.29%)23.58 (+37.29%)0.70 (-74.97%)
0a6c11c — 2025-12-04 00:39:1024.54 (n/a)22.73 (n/a)23.96 (n/a)17.18 (n/a)2.79 (n/a)

matrix_vector_mul_8192x2048_4_2col0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
331dcca — 2026-01-08 18:34:1025.08 (+1.32%)24.68 (+13.42%)24.61 (+3.26%)24.47 (+48.32%)0.24 (-93.56%)25.07 (+1.32%)24.67 (+13.42%)24.59 (+3.26%)24.46 (+48.32%)0.24 (-93.56%)
a4b6ffe — 2026-01-05 19:24:0524.76 (n/a)21.76 (n/a)23.83 (n/a)16.50 (n/a)3.74 (n/a)24.74 (n/a)21.75 (n/a)23.82 (n/a)16.49 (n/a)3.73 (n/a)

matrix_vector_mul_8192x2048_4_4col

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)
130b6ea — 2025-12-05 21:15:4940.94 (-2.19%)39.74 (-1.10%)39.76 (-0.60%)38.17 (-2.14%)1.04 (+4.39%)
0a6c11c — 2025-12-04 00:39:1041.86 (n/a)40.18 (n/a)40.00 (n/a)39.00 (n/a)1.00 (n/a)

matrix_vector_mul_8192x2048_4_4col0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
331dcca — 2026-01-08 18:34:1040.74 (-1.09%)37.16 (-7.15%)40.25 (+1.47%)24.51 (-37.44%)7.07 (+736.13%)40.72 (-1.09%)37.13 (-7.15%)40.23 (+1.47%)24.50 (-37.44%)7.07 (+736.13%)
a4b6ffe — 2026-01-05 19:24:0541.19 (n/a)40.02 (n/a)39.67 (n/a)39.18 (n/a)0.85 (n/a)41.17 (n/a)39.99 (n/a)39.64 (n/a)39.16 (n/a)0.85 (n/a)

matrix_vector_mul_8192x2048_4_8col

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)
130b6ea — 2025-12-05 21:15:4946.71 (-1.04%)44.91 (-2.34%)44.71 (-2.98%)43.80 (-1.44%)0.96 (+6.93%)
0a6c11c — 2025-12-04 00:39:1047.20 (n/a)45.98 (n/a)46.08 (n/a)44.44 (n/a)0.90 (n/a)

matrix_vector_mul_8192x2048_4_8col0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
331dcca — 2026-01-08 18:34:1046.28 (+5.68%)44.89 (+14.59%)44.74 (+4.26%)44.20 (+62.80%)0.84 (-87.96%)46.25 (+5.68%)44.86 (+14.59%)44.71 (+4.26%)44.17 (+62.80%)0.84 (-87.96%)
a4b6ffe — 2026-01-05 19:24:0543.79 (n/a)39.17 (n/a)42.91 (n/a)27.15 (n/a)6.98 (n/a)43.77 (n/a)39.15 (n/a)42.88 (n/a)27.13 (n/a)6.98 (n/a)

matrix_vector_mul_8192x2048_4tsi_1024tso_1col0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
116b247 — 2026-03-07 19:03:3813.32 (-0.18%)13.19 (-0.10%)13.16 (-0.20%)13.12 (-0.02%)0.08 (-1.34%)13.31 (-0.18%)13.18 (-0.10%)13.15 (-0.20%)13.11 (-0.02%)0.08 (-1.33%)
897d04e — 2026-03-06 22:12:2513.34 (n/a)13.20 (n/a)13.18 (n/a)13.12 (n/a)0.08 (n/a)13.33 (n/a)13.19 (n/a)13.17 (n/a)13.11 (n/a)0.08 (n/a)

matrix_vector_mul_8192x2048_4tsi_1024tso_2col0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
116b247 — 2026-03-07 19:03:3825.31 (+5.04%)24.28 (+5.25%)24.11 (+0.40%)23.73 (+23.24%)0.61 (-71.48%)25.29 (+5.04%)24.26 (+5.25%)24.09 (+0.40%)23.72 (+23.24%)0.61 (-71.48%)
897d04e — 2026-03-06 22:12:2524.10 (n/a)23.07 (n/a)24.01 (n/a)19.26 (n/a)2.13 (n/a)24.08 (n/a)23.05 (n/a)23.99 (n/a)19.25 (n/a)2.13 (n/a)

matrix_vector_mul_8192x2048_4tsi_1024tso_4col0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
116b247 — 2026-03-07 19:03:3841.90 (+2.69%)36.64 (-7.42%)39.17 (-1.47%)23.75 (-37.96%)7.30 (+566.47%)41.87 (+2.69%)36.62 (-7.42%)39.15 (-1.47%)23.73 (-37.96%)7.30 (+566.47%)
897d04e — 2026-03-06 22:12:2540.80 (n/a)39.58 (n/a)39.75 (n/a)38.28 (n/a)1.10 (n/a)40.78 (n/a)39.55 (n/a)39.73 (n/a)38.26 (n/a)1.10 (n/a)

matrix_vector_mul_8192x2048_4tsi_1024tso_8col0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
116b247 — 2026-03-07 19:03:3847.86 (-2.93%)46.71 (+1.27%)46.37 (+2.35%)45.41 (+1.76%)1.04 (-43.76%)47.83 (-2.93%)46.68 (+1.27%)46.34 (+2.35%)45.38 (+1.76%)1.04 (-43.76%)
897d04e — 2026-03-06 22:12:2549.31 (n/a)46.12 (n/a)45.30 (n/a)44.62 (n/a)1.85 (n/a)49.28 (n/a)46.10 (n/a)45.28 (n/a)44.59 (n/a)1.85 (n/a)

mem_copy_16_cores_2_chans_2048_tile_128_False

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
130b6ea — 2025-12-05 21:15:490.07 (+22.03%)0.06 (+27.86%)0.06 (+35.49%)0.05 (+15.00%)0.01 (+49.69%)173.30 (-13.00%)137.94 (-21.32%)133.60 (-26.19%)123.40 (-18.06%)18.40 (+8.53%)
0a6c11c — 2025-12-04 00:39:100.05 (n/a)0.05 (n/a)0.05 (n/a)0.04 (n/a)0.00 (n/a)199.20 (n/a)175.32 (n/a)181.00 (n/a)150.60 (n/a)16.96 (n/a)

mem_copy_16_cores_2_chans_2048_tile_128_False0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.08 (+2.84%)0.07 (+22.13%)0.08 (+27.90%)0.05 (+38.90%)0.02 (-13.16%)178.20 (-28.03%)116.16 (-21.71%)100.30 (-21.82%)97.30 (-2.80%)34.91 (-40.53%)
897d04e — 2026-03-06 22:12:250.08 (n/a)0.06 (n/a)0.06 (n/a)0.03 (n/a)0.02 (n/a)247.60 (n/a)148.38 (n/a)128.30 (n/a)100.10 (n/a)58.70 (n/a)

mem_copy_1_cols_1_channels_2048_tile_2048

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
003b49d — 2025-11-14 19:30:540.05 (-16.99%)0.05 (-12.47%)0.05 (-7.32%)0.04 (-11.17%)0.00 (-29.61%)197.00 (+12.57%)171.00 (+13.70%)164.00 (+7.89%)153.00 (+20.47%)17.49 (-4.32%)
0999a9a — 2025-11-13 19:24:080.06 (n/a)0.05 (n/a)0.05 (n/a)0.04 (n/a)0.01 (n/a)175.00 (n/a)150.40 (n/a)152.00 (n/a)127.00 (n/a)18.28 (n/a)

mem_copy_1_cols_2_channels_2048_tile_1024

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
003b49d — 2025-11-14 19:30:540.06 (+5.56%)0.05 (-1.59%)0.05 (-1.39%)0.03 (-14.07%)0.01 (+25.03%)263.00 (+16.37%)166.00 (+4.53%)144.00 (+1.41%)126.00 (-5.26%)49.38 (+43.17%)
0999a9a — 2025-11-13 19:24:080.06 (n/a)0.05 (n/a)0.05 (n/a)0.03 (n/a)0.01 (n/a)226.00 (n/a)158.80 (n/a)142.00 (n/a)133.00 (n/a)34.49 (n/a)

mem_copy_1_cores_1_chans_2048_tile_2048_False

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
130b6ea — 2025-12-05 21:15:490.08 (+10.93%)0.07 (+7.79%)0.07 (+3.55%)0.05 (-1.73%)0.01 (+36.73%)150.20 (+1.76%)121.68 (-6.36%)125.60 (-3.38%)97.90 (-9.85%)18.56 (+21.98%)
0a6c11c — 2025-12-04 00:39:100.08 (n/a)0.06 (n/a)0.06 (n/a)0.06 (n/a)0.01 (n/a)147.60 (n/a)129.94 (n/a)130.00 (n/a)108.60 (n/a)15.21 (n/a)

mem_copy_1_cores_1_chans_2048_tile_2048_False0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.11 (-1.27%)0.09 (+22.41%)0.09 (+15.53%)0.08 (+39.61%)0.01 (-38.24%)106.40 (-28.35%)87.96 (-22.17%)89.70 (-13.42%)75.10 (+1.35%)13.24 (-58.23%)
897d04e — 2026-03-06 22:12:250.11 (n/a)0.08 (n/a)0.08 (n/a)0.06 (n/a)0.02 (n/a)148.50 (n/a)113.02 (n/a)103.60 (n/a)74.10 (n/a)31.69 (n/a)

mem_copy_2_cols_1_channels_2048_tile_1024

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
003b49d — 2025-11-14 19:30:540.06 (-15.32%)0.05 (-17.09%)0.05 (-23.81%)0.04 (-15.54%)0.01 (-18.18%)193.00 (+18.40%)160.80 (+20.36%)168.00 (+31.25%)124.00 (+18.10%)22.92 (+11.39%)
0999a9a — 2025-11-13 19:24:080.07 (n/a)0.06 (n/a)0.06 (n/a)0.05 (n/a)0.01 (n/a)163.00 (n/a)133.60 (n/a)128.00 (n/a)105.00 (n/a)20.58 (n/a)

mem_copy_2_cols_2_channels_2048_tile_512

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
003b49d — 2025-11-14 19:30:540.06 (-23.02%)0.05 (-6.98%)0.05 (-2.13%)0.04 (-15.31%)0.01 (-33.29%)196.00 (+18.07%)147.00 (+6.52%)141.00 (+2.17%)126.00 (+29.90%)25.27 (+8.13%)
0999a9a — 2025-11-13 19:24:080.08 (n/a)0.06 (n/a)0.06 (n/a)0.05 (n/a)0.01 (n/a)166.00 (n/a)138.00 (n/a)138.00 (n/a)97.00 (n/a)23.37 (n/a)

mem_copy_2_cores_1_chans_2048_tile_1024_False

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
130b6ea — 2025-12-05 21:15:490.10 (+31.37%)0.08 (+29.12%)0.08 (+29.06%)0.07 (+41.80%)0.01 (+0.23%)109.90 (-29.46%)98.30 (-23.38%)100.80 (-22.52%)79.40 (-23.87%)10.54 (-46.17%)
0a6c11c — 2025-12-04 00:39:100.08 (n/a)0.07 (n/a)0.06 (n/a)0.05 (n/a)0.01 (n/a)155.80 (n/a)128.30 (n/a)130.10 (n/a)104.30 (n/a)19.58 (n/a)

mem_copy_2_cores_1_chans_2048_tile_1024_False0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.09 (-22.79%)0.08 (-17.44%)0.08 (-21.24%)0.05 (+5.00%)0.01 (-44.37%)154.20 (-4.76%)108.20 (+14.69%)96.90 (+27.00%)95.90 (+29.59%)25.72 (-32.33%)
897d04e — 2026-03-06 22:12:250.11 (n/a)0.09 (n/a)0.11 (n/a)0.05 (n/a)0.03 (n/a)161.90 (n/a)94.34 (n/a)76.30 (n/a)74.00 (n/a)38.01 (n/a)

mem_copy_2_cores_2_chans_2048_tile_1024_False

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
130b6ea — 2025-12-05 21:15:490.10 (+25.02%)0.07 (+10.01%)0.07 (+7.53%)0.05 (-1.99%)0.01 (+77.08%)150.10 (+2.04%)114.60 (-7.22%)111.70 (-6.99%)83.90 (-20.02%)21.23 (+43.54%)
0a6c11c — 2025-12-04 00:39:100.08 (n/a)0.07 (n/a)0.07 (n/a)0.06 (n/a)0.01 (n/a)147.10 (n/a)123.52 (n/a)120.10 (n/a)104.90 (n/a)14.79 (n/a)

mem_copy_2_cores_2_chans_2048_tile_1024_False0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.10 (+2.59%)0.08 (+2.66%)0.09 (-0.88%)0.06 (+44.19%)0.02 (-30.51%)139.80 (-30.65%)101.62 (-9.33%)91.60 (+0.88%)82.40 (-2.60%)22.95 (-54.23%)
897d04e — 2026-03-06 22:12:250.10 (n/a)0.08 (n/a)0.09 (n/a)0.04 (n/a)0.02 (n/a)201.60 (n/a)112.08 (n/a)90.80 (n/a)84.60 (n/a)50.14 (n/a)

mem_copy_4_cols_1_channels_2048_tile_512

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
003b49d — 2025-11-14 19:30:540.07 (+11.97%)0.05 (+3.09%)0.05 (+2.10%)0.03 (-11.35%)0.01 (+41.65%)229.00 (+12.81%)156.00 (-0.51%)143.00 (-2.05%)117.00 (-10.69%)38.21 (+48.71%)
0999a9a — 2025-11-13 19:24:080.06 (n/a)0.05 (n/a)0.05 (n/a)0.04 (n/a)0.01 (n/a)203.00 (n/a)156.80 (n/a)146.00 (n/a)131.00 (n/a)25.69 (n/a)

mem_copy_4_cols_2_channels_2048_tile_256

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
003b49d — 2025-11-14 19:30:540.06 (+9.02%)0.05 (+5.59%)0.05 (+0.00%)0.04 (+3.63%)0.01 (+25.48%)193.00 (-3.50%)147.40 (-4.66%)147.00 (+0.00%)122.00 (-8.27%)25.24 (+7.48%)
0999a9a — 2025-11-13 19:24:080.06 (n/a)0.05 (n/a)0.05 (n/a)0.04 (n/a)0.01 (n/a)200.00 (n/a)154.60 (n/a)147.00 (n/a)133.00 (n/a)23.48 (n/a)

mem_copy_4_cores_1_chans_2048_tile_512_False

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
130b6ea — 2025-12-05 21:15:490.09 (-3.98%)0.07 (+4.00%)0.07 (+7.04%)0.06 (+3.67%)0.01 (-15.76%)129.70 (-3.57%)113.30 (-4.50%)116.90 (-6.55%)90.40 (+4.15%)15.04 (-13.80%)
0a6c11c — 2025-12-04 00:39:100.09 (n/a)0.07 (n/a)0.07 (n/a)0.06 (n/a)0.01 (n/a)134.50 (n/a)118.64 (n/a)125.10 (n/a)86.80 (n/a)17.45 (n/a)

mem_copy_4_cores_1_chans_2048_tile_512_False0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.09 (-1.25%)0.07 (-10.46%)0.07 (-10.43%)0.04 (-27.35%)0.02 (+45.79%)201.30 (+37.59%)126.36 (+16.98%)110.80 (+11.58%)96.30 (+1.26%)43.39 (+101.85%)
897d04e — 2026-03-06 22:12:250.09 (n/a)0.08 (n/a)0.08 (n/a)0.06 (n/a)0.01 (n/a)146.30 (n/a)108.02 (n/a)99.30 (n/a)95.10 (n/a)21.49 (n/a)

mem_copy_4_cores_2_chans_2048_tile_512_False

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
130b6ea — 2025-12-05 21:15:490.08 (+9.57%)0.07 (+16.85%)0.07 (+19.58%)0.06 (+24.52%)0.01 (+1.05%)134.50 (-19.70%)116.64 (-14.92%)116.80 (-16.39%)98.00 (-8.75%)15.71 (-25.13%)
0a6c11c — 2025-12-04 00:39:100.08 (n/a)0.06 (n/a)0.06 (n/a)0.05 (n/a)0.01 (n/a)167.50 (n/a)137.10 (n/a)139.70 (n/a)107.40 (n/a)20.99 (n/a)

mem_copy_4_cores_2_chans_2048_tile_512_False0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.11 (+1.27%)0.09 (-4.02%)0.10 (+0.84%)0.05 (-4.62%)0.02 (+9.04%)149.70 (+4.83%)96.78 (+5.26%)78.40 (-0.76%)76.70 (-1.29%)31.30 (+9.81%)
897d04e — 2026-03-06 22:12:250.11 (n/a)0.09 (n/a)0.10 (n/a)0.06 (n/a)0.02 (n/a)142.80 (n/a)91.94 (n/a)79.00 (n/a)77.70 (n/a)28.50 (n/a)

mem_copy_8_cols_1_channels_2048_tile_256

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
003b49d — 2025-11-14 19:30:540.05 (-13.48%)0.05 (-17.79%)0.05 (-15.33%)0.03 (-25.09%)0.01 (+6.67%)267.00 (+33.50%)176.80 (+24.51%)150.00 (+18.11%)141.00 (+15.57%)47.41 (+61.41%)
0999a9a — 2025-11-13 19:24:080.06 (n/a)0.06 (n/a)0.06 (n/a)0.04 (n/a)0.01 (n/a)200.00 (n/a)142.00 (n/a)127.00 (n/a)122.00 (n/a)29.37 (n/a)

mem_copy_8_cols_2_channels_2048_tile_128

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
003b49d — 2025-11-14 19:30:540.05 (+13.38%)0.05 (+23.85%)0.05 (+37.91%)0.05 (+30.72%)0.00 (-56.85%)166.00 (-23.50%)153.40 (-20.60%)153.00 (-27.49%)142.00 (-11.80%)7.63 (-70.60%)
0999a9a — 2025-11-13 19:24:080.05 (n/a)0.04 (n/a)0.04 (n/a)0.04 (n/a)0.01 (n/a)217.00 (n/a)193.20 (n/a)211.00 (n/a)161.00 (n/a)25.96 (n/a)

mem_copy_8_cores_1_chans_2048_tile_256_False

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
130b6ea — 2025-12-05 21:15:490.09 (+16.31%)0.08 (+15.34%)0.08 (+14.60%)0.07 (+19.30%)0.01 (-5.41%)117.30 (-16.15%)107.80 (-13.61%)108.50 (-12.78%)94.10 (-13.99%)8.29 (-32.01%)
0a6c11c — 2025-12-04 00:39:100.07 (n/a)0.07 (n/a)0.07 (n/a)0.06 (n/a)0.01 (n/a)139.90 (n/a)124.78 (n/a)124.40 (n/a)109.40 (n/a)12.20 (n/a)

mem_copy_8_cores_1_chans_2048_tile_256_False0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.09 (+8.86%)0.07 (-9.81%)0.07 (-10.03%)0.05 (-24.06%)0.02 (+94.54%)180.10 (+31.75%)127.80 (+15.51%)116.40 (+11.17%)94.60 (-8.16%)34.33 (+135.34%)
897d04e — 2026-03-06 22:12:250.08 (n/a)0.07 (n/a)0.08 (n/a)0.06 (n/a)0.01 (n/a)136.70 (n/a)110.64 (n/a)104.70 (n/a)103.00 (n/a)14.59 (n/a)

mem_copy_8_cores_2_chans_2048_tile_256_False

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
130b6ea — 2025-12-05 21:15:490.09 (+12.61%)0.07 (-2.24%)0.07 (-8.85%)0.06 (+5.13%)0.01 (+35.50%)136.50 (-4.88%)120.86 (+2.96%)125.10 (+9.64%)91.60 (-11.15%)15.72 (+10.46%)
0a6c11c — 2025-12-04 00:39:100.08 (n/a)0.07 (n/a)0.07 (n/a)0.06 (n/a)0.01 (n/a)143.50 (n/a)117.38 (n/a)114.10 (n/a)103.10 (n/a)14.23 (n/a)

mem_copy_8_cores_2_chans_2048_tile_256_False0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.08 (-1.14%)0.07 (+9.06%)0.08 (+0.36%)0.05 (+22.18%)0.01 (-34.77%)151.70 (-18.13%)112.54 (-11.89%)101.70 (-0.29%)101.50 (+1.10%)21.98 (-43.16%)
897d04e — 2026-03-06 22:12:250.08 (n/a)0.07 (n/a)0.08 (n/a)0.04 (n/a)0.02 (n/a)185.30 (n/a)127.72 (n/a)102.00 (n/a)100.40 (n/a)38.66 (n/a)

mha

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
130b6ea — 2025-12-05 21:15:490.18 (+0.16%)0.18 (+0.19%)0.18 (+0.12%)0.18 (+0.32%)0.00 (-52.24%)47253.90 (-0.32%)47213.70 (-0.19%)47210.90 (-0.12%)47167.70 (-0.16%)28.71 (-52.47%)
0a6c11c — 2025-12-04 00:39:100.18 (n/a)0.18 (n/a)0.18 (n/a)0.18 (n/a)0.00 (n/a)47405.00 (n/a)47301.72 (n/a)47265.90 (n/a)47241.80 (n/a)60.40 (n/a)

mha0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
84d3478 — 2026-02-17 23:25:130.18 (-0.51%)0.18 (-0.36%)0.18 (-0.41%)0.18 (-0.21%)0.00 (-35.21%)47468.60 (+0.21%)47403.94 (+0.36%)47426.00 (+0.41%)47314.40 (+0.51%)68.83 (-34.73%)
1531781 — 2026-01-23 18:49:240.18 (n/a)0.18 (n/a)0.18 (n/a)0.18 (n/a)0.00 (n/a)47367.50 (n/a)47232.68 (n/a)47232.00 (n/a)47073.00 (n/a)105.45 (n/a)

mha_16384_64_1_8_0_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.18 (+0.04%)0.18 (+0.11%)0.18 (+0.01%)0.18 (+0.29%)0.00 (-41.83%)47397.40 (-0.29%)47329.26 (-0.11%)47334.60 (-0.01%)47248.90 (-0.04%)64.10 (-42.04%)
897d04e — 2026-03-06 22:12:250.18 (n/a)0.18 (n/a)0.18 (n/a)0.18 (n/a)0.00 (n/a)47533.70 (n/a)47380.56 (n/a)47337.90 (n/a)47268.00 (n/a)110.60 (n/a)

relu_1_cols_1_channels_2048_tile_2048

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.11 (+24.27%)0.08 (+9.32%)0.08 (-7.20%)0.05 (-12.04%)0.03 (+51.55%)171.70 (+13.71%)111.32 (-3.74%)100.60 (+7.71%)74.10 (-19.46%)41.87 (+32.67%)
897d04e — 2026-03-06 22:12:250.09 (n/a)0.07 (n/a)0.09 (n/a)0.05 (n/a)0.02 (n/a)151.00 (n/a)115.64 (n/a)93.40 (n/a)92.00 (n/a)31.56 (n/a)

relu_2_cols_1_channels_2048_tile_1024

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.09 (-3.09%)0.08 (-5.30%)0.08 (-3.09%)0.05 (-27.13%)0.02 (+57.81%)179.40 (+37.26%)114.20 (+10.19%)97.20 (+3.18%)96.30 (+3.22%)36.51 (+127.33%)
897d04e — 2026-03-06 22:12:250.09 (n/a)0.08 (n/a)0.09 (n/a)0.06 (n/a)0.01 (n/a)130.70 (n/a)103.64 (n/a)94.20 (n/a)93.30 (n/a)16.06 (n/a)

relu_4_cols_1_channels_2048_tile_512

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.09 (-19.79%)0.08 (-17.09%)0.09 (-18.30%)0.04 (-9.44%)0.02 (-25.69%)190.70 (+10.42%)114.68 (+17.74%)95.90 (+22.48%)94.20 (+24.60%)42.51 (+0.78%)
897d04e — 2026-03-06 22:12:250.11 (n/a)0.09 (n/a)0.10 (n/a)0.05 (n/a)0.03 (n/a)172.70 (n/a)97.40 (n/a)78.30 (n/a)75.60 (n/a)42.18 (n/a)

relu_8_cols_1_channels_2048_tile_256

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.08 (+4.84%)0.07 (+3.16%)0.08 (-1.09%)0.05 (+14.59%)0.02 (-10.29%)176.70 (-12.74%)116.60 (-5.23%)104.50 (+1.16%)96.70 (-4.54%)33.79 (-23.97%)
897d04e — 2026-03-06 22:12:250.08 (n/a)0.07 (n/a)0.08 (n/a)0.04 (n/a)0.02 (n/a)202.50 (n/a)123.04 (n/a)103.30 (n/a)101.30 (n/a)44.44 (n/a)

rms_norm_1_cols_1_channels_2048_tile_2048

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.09 (+0.81%)0.08 (+1.40%)0.08 (-0.65%)0.06 (+4.69%)0.01 (-8.71%)136.50 (-4.48%)107.22 (-1.85%)98.00 (+0.72%)95.40 (-0.83%)17.50 (-13.05%)
897d04e — 2026-03-06 22:12:250.09 (n/a)0.08 (n/a)0.08 (n/a)0.06 (n/a)0.01 (n/a)142.90 (n/a)109.24 (n/a)97.30 (n/a)96.20 (n/a)20.13 (n/a)

rms_norm_1_cols_2_channels_2048_tile_1024

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.11 (+24.45%)0.08 (+4.31%)0.08 (+1.46%)0.04 (-31.47%)0.03 (+122.17%)214.20 (+45.91%)115.84 (+7.14%)96.80 (-1.43%)77.90 (-19.61%)56.73 (+161.44%)
897d04e — 2026-03-06 22:12:250.08 (n/a)0.08 (n/a)0.08 (n/a)0.06 (n/a)0.01 (n/a)146.80 (n/a)108.12 (n/a)98.20 (n/a)96.90 (n/a)21.70 (n/a)

rms_norm_2_cols_1_channels_2048_tile_1024

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.10 (-15.23%)0.08 (-17.17%)0.08 (-20.33%)0.07 (+1.01%)0.01 (-37.51%)123.90 (-1.04%)100.48 (+17.74%)97.90 (+25.51%)80.30 (+17.91%)15.65 (-30.99%)
897d04e — 2026-03-06 22:12:250.12 (n/a)0.10 (n/a)0.10 (n/a)0.07 (n/a)0.02 (n/a)125.20 (n/a)85.34 (n/a)78.00 (n/a)68.10 (n/a)22.68 (n/a)

rms_norm_2_cols_2_channels_2048_tile_512

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.11 (+29.22%)0.08 (-1.05%)0.09 (+1.51%)0.05 (-8.49%)0.02 (+104.00%)153.00 (+9.29%)112.96 (+6.57%)96.00 (-1.54%)74.90 (-22.54%)34.30 (+80.39%)
897d04e — 2026-03-06 22:12:250.08 (n/a)0.08 (n/a)0.08 (n/a)0.06 (n/a)0.01 (n/a)140.00 (n/a)106.00 (n/a)97.50 (n/a)96.70 (n/a)19.02 (n/a)

rms_norm_4_cols_1_channels_2048_tile_512

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.08 (-5.57%)0.07 (-14.65%)0.08 (-4.46%)0.04 (-26.93%)0.02 (+37.74%)208.20 (+36.79%)131.30 (+23.56%)100.10 (+4.71%)98.80 (+5.89%)48.25 (+87.68%)
897d04e — 2026-03-06 22:12:250.09 (n/a)0.08 (n/a)0.09 (n/a)0.05 (n/a)0.01 (n/a)152.20 (n/a)106.26 (n/a)95.60 (n/a)93.30 (n/a)25.71 (n/a)

rms_norm_4_cols_2_channels_2048_tile_256

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.10 (+20.71%)0.09 (+6.07%)0.09 (+2.19%)0.06 (-6.39%)0.02 (+67.12%)139.30 (+6.83%)99.70 (-3.65%)95.40 (-2.15%)79.30 (-17.14%)23.11 (+53.25%)
897d04e — 2026-03-06 22:12:250.09 (n/a)0.08 (n/a)0.08 (n/a)0.06 (n/a)0.01 (n/a)130.40 (n/a)103.48 (n/a)97.50 (n/a)95.70 (n/a)15.08 (n/a)

rms_norm_8_cols_1_channels_2048_tile_256

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.08 (-8.55%)0.07 (-5.95%)0.08 (-3.69%)0.05 (+20.52%)0.01 (-29.58%)157.90 (-17.03%)122.44 (+2.51%)107.80 (+3.85%)104.10 (+9.35%)23.78 (-40.25%)
897d04e — 2026-03-06 22:12:250.09 (n/a)0.07 (n/a)0.08 (n/a)0.04 (n/a)0.02 (n/a)190.30 (n/a)119.44 (n/a)103.80 (n/a)95.20 (n/a)39.79 (n/a)

rms_norm_8_cols_2_channels_2048_tile_128

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.10 (-4.69%)0.09 (+0.22%)0.10 (-0.93%)0.05 (+21.82%)0.02 (-19.28%)161.60 (-17.93%)99.12 (-5.51%)83.80 (+0.84%)81.60 (+4.88%)34.96 (-32.11%)
897d04e — 2026-03-06 22:12:250.11 (n/a)0.09 (n/a)0.10 (n/a)0.04 (n/a)0.03 (n/a)196.90 (n/a)104.90 (n/a)83.10 (n/a)77.80 (n/a)51.49 (n/a)

rope_1_cols_2_channels_4096_tile_4096_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
1b2d230 — 2026-01-13 22:38:030.19 (-21.66%)0.17 (-8.53%)0.18 (+2.09%)0.14 (-6.71%)0.02 (-38.07%)174.10 (+7.20%)150.06 (+8.11%)139.60 (-2.04%)129.80 (+27.63%)20.36 (-10.65%)
d48746f — 2026-01-12 22:17:100.24 (n/a)0.18 (n/a)0.17 (n/a)0.15 (n/a)0.04 (n/a)162.40 (n/a)138.80 (n/a)142.50 (n/a)101.70 (n/a)22.79 (n/a)

rope_1c_32rows_512cols_32arows_0m

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:381.11 (+4.28%)0.96 (+8.58%)1.07 (+17.68%)0.57 (-20.02%)0.23 (+54.42%)172.90 (+25.02%)108.96 (-4.03%)91.80 (-15.08%)88.80 (-4.10%)36.04 (+87.91%)
897d04e — 2026-03-06 22:12:251.06 (n/a)0.89 (n/a)0.91 (n/a)0.71 (n/a)0.15 (n/a)138.30 (n/a)113.54 (n/a)108.10 (n/a)92.60 (n/a)19.18 (n/a)

rope_1c_32rows_512cols_8arows_0m

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.50 (-19.01%)0.47 (-6.08%)0.48 (-0.63%)0.43 (-3.42%)0.03 (-59.90%)171.60 (+3.56%)156.58 (+5.50%)153.30 (+0.66%)147.80 (+23.48%)9.13 (-46.61%)
897d04e — 2026-03-06 22:12:250.62 (n/a)0.50 (n/a)0.48 (n/a)0.44 (n/a)0.07 (n/a)165.70 (n/a)148.42 (n/a)152.30 (n/a)119.70 (n/a)17.10 (n/a)

rope_2_cols_2_channels_4096_tile_2048_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
1b2d230 — 2026-01-13 22:38:030.21 (-0.56%)0.17 (-10.62%)0.16 (-13.58%)0.13 (-13.33%)0.03 (+17.05%)183.70 (+15.39%)151.76 (+12.77%)155.20 (+15.73%)117.30 (+0.60%)24.24 (+35.73%)
d48746f — 2026-01-12 22:17:100.21 (n/a)0.19 (n/a)0.18 (n/a)0.15 (n/a)0.02 (n/a)159.20 (n/a)134.58 (n/a)134.10 (n/a)116.60 (n/a)17.86 (n/a)

rope_2c_32rows_512cols_32arows_0m

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:381.17 (+8.22%)0.99 (+10.99%)1.14 (+22.95%)0.53 (-6.80%)0.27 (+35.22%)185.10 (+7.30%)108.36 (-6.30%)86.20 (-18.60%)84.10 (-7.58%)43.57 (+31.68%)
897d04e — 2026-03-06 22:12:251.08 (n/a)0.90 (n/a)0.93 (n/a)0.57 (n/a)0.20 (n/a)172.50 (n/a)115.64 (n/a)105.90 (n/a)91.00 (n/a)33.09 (n/a)

rope_2c_32rows_512cols_8arows_0m

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.86 (-1.88%)0.63 (+0.72%)0.55 (+7.29%)0.54 (+10.25%)0.14 (-21.79%)137.40 (-9.31%)120.16 (-3.14%)133.90 (-6.82%)85.80 (+1.90%)22.35 (-27.83%)
897d04e — 2026-03-06 22:12:250.88 (n/a)0.63 (n/a)0.51 (n/a)0.49 (n/a)0.18 (n/a)151.50 (n/a)124.06 (n/a)143.70 (n/a)84.20 (n/a)30.96 (n/a)

rope_4_cols_2_channels_4096_tile_1024_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
1b2d230 — 2026-01-13 22:38:030.22 (+18.92%)0.17 (+6.24%)0.17 (+5.57%)0.14 (-0.50%)0.03 (+90.77%)171.10 (+0.47%)147.40 (-4.39%)148.10 (-5.31%)110.10 (-15.89%)23.59 (+60.67%)
d48746f — 2026-01-12 22:17:100.19 (n/a)0.16 (n/a)0.16 (n/a)0.14 (n/a)0.02 (n/a)170.30 (n/a)154.16 (n/a)156.40 (n/a)130.90 (n/a)14.69 (n/a)

rope_8_cols_2_channels_4096_tile_512_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
1b2d230 — 2026-01-13 22:38:030.17 (-1.30%)0.15 (+6.63%)0.15 (-8.78%)0.14 (+39.68%)0.01 (-67.35%)174.90 (-28.41%)163.42 (-10.36%)167.40 (+9.63%)146.50 (+1.31%)11.19 (-76.17%)
d48746f — 2026-01-12 22:17:100.17 (n/a)0.14 (n/a)0.16 (n/a)0.10 (n/a)0.03 (n/a)244.30 (n/a)182.30 (n/a)152.70 (n/a)144.60 (n/a)46.98 (n/a)

rope_8c_32rows_512cols_32arows_0m

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:381.12 (+5.41%)1.01 (+14.61%)1.12 (+8.86%)0.60 (+12.44%)0.23 (-0.80%)162.60 (-11.05%)103.04 (-13.54%)88.10 (-8.13%)87.50 (-5.20%)33.30 (-14.16%)
897d04e — 2026-03-06 22:12:251.07 (n/a)0.88 (n/a)1.03 (n/a)0.54 (n/a)0.23 (n/a)182.80 (n/a)119.18 (n/a)95.90 (n/a)92.30 (n/a)38.79 (n/a)

rope_8c_32rows_512cols_8arows_0m

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.68 (+20.83%)0.54 (+11.73%)0.48 (-8.69%)0.37 (+1.30%)0.14 (+56.42%)200.40 (-1.28%)145.26 (-8.24%)153.80 (+9.54%)108.70 (-17.21%)38.15 (+21.58%)
897d04e — 2026-03-06 22:12:250.56 (n/a)0.48 (n/a)0.52 (n/a)0.36 (n/a)0.09 (n/a)203.00 (n/a)158.30 (n/a)140.40 (n/a)131.30 (n/a)31.38 (n/a)

sigmoid_1_cols_1_channels_2048_tile_2048

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.09 (-0.46%)0.07 (-4.80%)0.07 (-18.35%)0.06 (+2.65%)0.02 (-9.43%)147.40 (-2.58%)118.84 (+4.23%)115.30 (+22.53%)94.00 (+0.53%)25.41 (-9.39%)
897d04e — 2026-03-06 22:12:250.09 (n/a)0.08 (n/a)0.09 (n/a)0.05 (n/a)0.02 (n/a)151.30 (n/a)114.02 (n/a)94.10 (n/a)93.50 (n/a)28.04 (n/a)

sigmoid_2_cols_1_channels_2048_tile_1024

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.09 (-20.06%)0.08 (-10.41%)0.08 (-22.91%)0.06 (+22.30%)0.01 (-58.61%)144.30 (-18.24%)108.54 (+1.63%)99.80 (+29.78%)95.60 (+25.13%)20.47 (-54.32%)
897d04e — 2026-03-06 22:12:250.11 (n/a)0.09 (n/a)0.11 (n/a)0.05 (n/a)0.03 (n/a)176.50 (n/a)106.80 (n/a)76.90 (n/a)76.40 (n/a)44.81 (n/a)

sigmoid_4_cols_1_channels_2048_tile_512

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.09 (+0.12%)0.08 (+4.09%)0.08 (+0.26%)0.06 (+21.14%)0.01 (-24.28%)141.70 (-17.42%)105.56 (-6.52%)100.00 (-0.20%)90.20 (-0.11%)20.60 (-38.11%)
897d04e — 2026-03-06 22:12:250.09 (n/a)0.08 (n/a)0.08 (n/a)0.05 (n/a)0.02 (n/a)171.60 (n/a)112.92 (n/a)100.20 (n/a)90.30 (n/a)33.29 (n/a)

sigmoid_8_cols_1_channels_2048_tile_256

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.08 (-20.36%)0.07 (-17.65%)0.08 (-19.82%)0.04 (+9.93%)0.02 (-39.67%)193.90 (-9.01%)122.40 (+12.11%)104.10 (+24.82%)103.80 (+25.67%)39.98 (-31.17%)
897d04e — 2026-03-06 22:12:250.10 (n/a)0.09 (n/a)0.10 (n/a)0.04 (n/a)0.03 (n/a)213.10 (n/a)109.18 (n/a)83.40 (n/a)82.60 (n/a)58.09 (n/a)

silu_1_cols_1_channels_2048_tile_2048

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.09 (-17.72%)0.07 (-11.34%)0.09 (-16.62%)0.05 (+16.70%)0.02 (-39.03%)161.60 (-14.32%)117.24 (+3.94%)94.30 (+19.97%)92.10 (+21.66%)33.22 (-34.92%)
897d04e — 2026-03-06 22:12:250.11 (n/a)0.08 (n/a)0.10 (n/a)0.04 (n/a)0.03 (n/a)188.60 (n/a)112.80 (n/a)78.60 (n/a)75.70 (n/a)51.04 (n/a)

silu_2_cols_1_channels_2048_tile_1024

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.09 (-7.06%)0.08 (+2.01%)0.09 (+14.93%)0.05 (-13.33%)0.02 (+3.32%)166.60 (+15.37%)111.16 (-0.73%)94.70 (-12.96%)93.70 (+7.58%)31.46 (+30.72%)
897d04e — 2026-03-06 22:12:250.09 (n/a)0.08 (n/a)0.08 (n/a)0.06 (n/a)0.02 (n/a)144.40 (n/a)111.98 (n/a)108.80 (n/a)87.10 (n/a)24.06 (n/a)

silu_4_cols_1_channels_2048_tile_512

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.08 (-6.94%)0.08 (-0.82%)0.08 (-0.16%)0.06 (+12.50%)0.01 (-24.00%)147.50 (-11.09%)109.12 (-1.39%)99.20 (+0.20%)96.50 (+7.46%)21.76 (-30.16%)
897d04e — 2026-03-06 22:12:250.09 (n/a)0.08 (n/a)0.08 (n/a)0.05 (n/a)0.02 (n/a)165.90 (n/a)110.66 (n/a)99.00 (n/a)89.80 (n/a)31.15 (n/a)

silu_8_cols_1_channels_2048_tile_256

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.08 (-20.33%)0.07 (-21.02%)0.08 (-16.69%)0.04 (-21.26%)0.02 (-18.16%)232.20 (+27.02%)133.34 (+27.09%)104.70 (+20.07%)102.20 (+25.40%)55.91 (+28.12%)
897d04e — 2026-03-06 22:12:250.10 (n/a)0.09 (n/a)0.09 (n/a)0.04 (n/a)0.02 (n/a)182.80 (n/a)104.92 (n/a)87.20 (n/a)81.50 (n/a)43.64 (n/a)

silu_mul_1_cols_2_channels_2048_tile_2048

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.15 (n/a)0.13 (n/a)0.14 (n/a)0.08 (n/a)0.03 (n/a)157.30 (n/a)100.84 (n/a)86.40 (n/a)83.90 (n/a)31.76 (n/a)

silu_mul_2_cols_2_channels_2048_tile_1024

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.13 (n/a)0.11 (n/a)0.12 (n/a)0.07 (n/a)0.02 (n/a)165.80 (n/a)112.68 (n/a)102.10 (n/a)95.50 (n/a)29.93 (n/a)

silu_mul_4_cols_2_channels_2048_tile_512

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.15 (n/a)0.13 (n/a)0.14 (n/a)0.09 (n/a)0.02 (n/a)135.10 (n/a)95.12 (n/a)85.70 (n/a)83.70 (n/a)22.37 (n/a)

silu_mul_8_cols_2_channels_2048_tile_256

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.14 (n/a)0.13 (n/a)0.14 (n/a)0.08 (n/a)0.03 (n/a)149.90 (n/a)102.32 (n/a)90.20 (n/a)85.00 (n/a)27.05 (n/a)

softmax_1_cols_2_channels_4096_tile_2048

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.20 (+5.50%)0.17 (+2.94%)0.19 (+15.09%)0.10 (-30.81%)0.04 (+88.24%)167.40 (+44.56%)102.30 (+3.06%)85.70 (-13.08%)80.50 (-5.29%)36.80 (+169.54%)
897d04e — 2026-03-06 22:12:250.19 (n/a)0.17 (n/a)0.17 (n/a)0.14 (n/a)0.02 (n/a)115.80 (n/a)99.26 (n/a)98.60 (n/a)85.00 (n/a)13.65 (n/a)

softmax_2_cols_2_channels_4096_tile_1024

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.20 (+3.04%)0.16 (-1.34%)0.19 (+12.95%)0.09 (-18.40%)0.05 (+70.25%)178.60 (+22.58%)114.22 (+9.11%)84.10 (-11.47%)81.40 (-2.86%)44.94 (+85.97%)
897d04e — 2026-03-06 22:12:250.20 (n/a)0.16 (n/a)0.17 (n/a)0.11 (n/a)0.03 (n/a)145.70 (n/a)104.68 (n/a)95.00 (n/a)83.80 (n/a)24.16 (n/a)

softmax_2_cols_2_channels_4096_tile_512

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.20 (-21.51%)0.17 (-3.26%)0.18 (+3.51%)0.11 (-6.45%)0.04 (-25.67%)154.00 (+6.87%)100.84 (+1.88%)89.00 (-3.37%)81.90 (+27.37%)30.32 (+2.50%)
897d04e — 2026-03-06 22:12:250.25 (n/a)0.18 (n/a)0.18 (n/a)0.11 (n/a)0.05 (n/a)144.10 (n/a)98.98 (n/a)92.10 (n/a)64.30 (n/a)29.58 (n/a)

swiglu

No metrics available.

swiglu_decode_1x2048x2048

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
130b6ea — 2025-12-05 21:15:490.00 (-15.46%)0.00 (-27.20%)0.00 (-28.17%)0.00 (-33.90%)0.00 (+1834.83%)5187.20 (+51.28%)4720.84 (+38.38%)4750.90 (+39.22%)4015.10 (+18.29%)390.23 (+3298.45%)
0a6c11c — 2025-12-04 00:39:100.00 (n/a)0.00 (n/a)0.00 (n/a)0.00 (n/a)0.00 (n/a)3428.80 (n/a)3411.62 (n/a)3412.60 (n/a)3394.20 (n/a)11.48 (n/a)

swiglu_decode_1x2048x2048_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.00 (+120.08%)0.00 (+132.35%)0.00 (+115.79%)0.00 (+225.05%)0.00 (-85.34%)1590.80 (-69.23%)1574.80 (-58.14%)1579.60 (-53.66%)1548.20 (-54.56%)16.11 (-97.95%)
897d04e — 2026-03-06 22:12:250.00 (n/a)0.00 (n/a)0.00 (n/a)0.00 (n/a)0.00 (n/a)5170.80 (n/a)3761.80 (n/a)3408.60 (n/a)3407.30 (n/a)787.66 (n/a)

swiglu_fused_decode_2048x2048_0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.01 (n/a)0.01 (n/a)0.01 (n/a)0.01 (n/a)0.00 (n/a)677.70 (n/a)668.98 (n/a)667.40 (n/a)658.80 (n/a)7.76 (n/a)

tanh_1_cols_1_channels_2048_tile_2048

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.08 (-2.31%)0.07 (-8.99%)0.08 (+6.85%)0.04 (-32.60%)0.02 (+102.00%)192.50 (+48.42%)124.20 (+16.60%)98.30 (-6.38%)97.60 (+2.41%)41.35 (+196.40%)
897d04e — 2026-03-06 22:12:250.09 (n/a)0.08 (n/a)0.08 (n/a)0.06 (n/a)0.01 (n/a)129.70 (n/a)106.52 (n/a)105.00 (n/a)95.30 (n/a)13.95 (n/a)

tanh_2_cols_1_channels_2048_tile_1024

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.09 (-21.98%)0.08 (-19.05%)0.08 (-24.70%)0.05 (+4.27%)0.01 (-45.80%)153.40 (-4.07%)111.16 (+17.11%)101.10 (+32.85%)95.70 (+28.11%)24.32 (-33.79%)
897d04e — 2026-03-06 22:12:250.11 (n/a)0.09 (n/a)0.11 (n/a)0.05 (n/a)0.02 (n/a)159.90 (n/a)94.92 (n/a)76.10 (n/a)74.70 (n/a)36.73 (n/a)

tanh_4_cols_1_channels_2048_tile_512

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.08 (+4.18%)0.07 (+0.60%)0.08 (+4.34%)0.05 (-23.35%)0.01 (+78.80%)168.30 (+30.47%)115.48 (+2.56%)102.20 (-4.13%)97.20 (-4.05%)29.91 (+130.65%)
897d04e — 2026-03-06 22:12:250.08 (n/a)0.07 (n/a)0.08 (n/a)0.06 (n/a)0.01 (n/a)129.00 (n/a)112.60 (n/a)106.60 (n/a)101.30 (n/a)12.97 (n/a)

tanh_8_cols_1_channels_2048_tile_256

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.08 (-1.94%)0.07 (-4.79%)0.08 (-2.70%)0.05 (-14.63%)0.01 (+31.48%)156.60 (+17.13%)117.10 (+6.45%)104.80 (+2.85%)101.30 (+2.01%)23.03 (+58.95%)
897d04e — 2026-03-06 22:12:250.08 (n/a)0.08 (n/a)0.08 (n/a)0.06 (n/a)0.01 (n/a)133.70 (n/a)110.00 (n/a)101.90 (n/a)99.30 (n/a)14.49 (n/a)

transpose_2048_M_64_N_1_cols_1_channels_64_m_64_n_8_s

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
130b6ea — 2025-12-05 21:15:493.93 (-7.02%)3.68 (+2.48%)3.66 (+1.82%)3.45 (+15.62%)0.15 (-63.64%)151.80 (-13.50%)142.70 (-3.59%)143.40 (-1.78%)133.50 (+7.49%)5.90 (-66.22%)
0a6c11c — 2025-12-04 00:39:104.22 (n/a)3.59 (n/a)3.59 (n/a)2.99 (n/a)0.42 (n/a)175.50 (n/a)148.02 (n/a)146.00 (n/a)124.20 (n/a)17.46 (n/a)

transpose_2048_M_64_N_1_cols_1_channels_64_m_64_n_8_s0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:384.07 (-8.19%)3.54 (-4.91%)3.60 (-10.57%)2.74 (-7.50%)0.55 (-15.23%)191.00 (+8.09%)151.18 (+4.65%)145.80 (+11.81%)129.00 (+8.95%)25.66 (-3.50%)
897d04e — 2026-03-06 22:12:254.43 (n/a)3.73 (n/a)4.02 (n/a)2.97 (n/a)0.65 (n/a)176.70 (n/a)144.46 (n/a)130.40 (n/a)118.40 (n/a)26.60 (n/a)

transpose_2048_M_64_N_1_cols_2_channels_64_m_64_n_8_s

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
130b6ea — 2025-12-05 21:15:495.11 (+15.57%)3.85 (-3.32%)3.40 (-18.64%)3.29 (+7.09%)0.71 (+49.14%)159.60 (-6.61%)140.18 (+4.80%)154.20 (+22.97%)102.60 (-13.49%)22.70 (+19.77%)
0a6c11c — 2025-12-04 00:39:104.42 (n/a)3.99 (n/a)4.18 (n/a)3.07 (n/a)0.48 (n/a)170.90 (n/a)133.76 (n/a)125.40 (n/a)118.60 (n/a)18.95 (n/a)

transpose_2048_M_64_N_1_cols_2_channels_64_m_64_n_8_s0

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:385.27 (-0.03%)4.14 (-4.76%)4.38 (-1.51%)2.53 (-9.07%)1.00 (+1.81%)207.40 (+9.97%)134.52 (+5.94%)119.70 (+1.53%)99.40 (+0.00%)42.00 (+16.62%)
897d04e — 2026-03-06 22:12:255.27 (n/a)4.35 (n/a)4.45 (n/a)2.78 (n/a)0.98 (n/a)188.60 (n/a)126.98 (n/a)117.90 (n/a)99.40 (n/a)36.02 (n/a)

weighted_rms_norm_1_cols_2_channels_2048_weights_2048

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.08 (+1.09%)0.08 (+8.10%)0.08 (+13.10%)0.05 (+4.38%)0.01 (+3.21%)162.20 (-4.19%)112.18 (-7.43%)99.80 (-11.60%)98.80 (-1.10%)27.97 (-1.90%)
897d04e — 2026-03-06 22:12:250.08 (n/a)0.07 (n/a)0.07 (n/a)0.05 (n/a)0.01 (n/a)169.30 (n/a)121.18 (n/a)112.90 (n/a)99.90 (n/a)28.51 (n/a)

weighted_rms_norm_2_cols_2_channels_2048_weights_1024

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.11 (+1.30%)0.08 (-12.60%)0.10 (-0.77%)0.04 (-28.74%)0.03 (+48.27%)196.70 (+40.30%)113.70 (+25.22%)79.40 (+0.76%)76.90 (-1.28%)53.14 (+92.40%)
897d04e — 2026-03-06 22:12:250.11 (n/a)0.10 (n/a)0.10 (n/a)0.06 (n/a)0.02 (n/a)140.20 (n/a)90.80 (n/a)78.80 (n/a)77.90 (n/a)27.62 (n/a)

weighted_rms_norm_4_cols_2_channels_2048_weights_512

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.11 (-0.30%)0.09 (-8.65%)0.09 (-11.78%)0.05 (-22.28%)0.02 (+30.37%)163.00 (+28.65%)101.50 (+13.81%)90.30 (+13.30%)77.40 (+0.26%)35.44 (+67.99%)
897d04e — 2026-03-06 22:12:250.11 (n/a)0.10 (n/a)0.10 (n/a)0.06 (n/a)0.02 (n/a)126.70 (n/a)89.18 (n/a)79.70 (n/a)77.20 (n/a)21.10 (n/a)

weighted_rms_norm_8_cols_2_channels_2048_weights_256

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
116b247 — 2026-03-07 19:03:380.09 (+6.93%)0.07 (-0.15%)0.08 (-0.72%)0.04 (-11.46%)0.02 (+21.64%)200.80 (+12.94%)122.86 (+2.88%)104.70 (+0.67%)96.20 (-6.51%)43.79 (+33.82%)
897d04e — 2026-03-06 22:12:250.08 (n/a)0.07 (n/a)0.08 (n/a)0.05 (n/a)0.01 (n/a)177.80 (n/a)119.42 (n/a)104.00 (n/a)102.90 (n/a)32.72 (n/a)

@github-actions
Copy link
Contributor

github-actions bot commented Mar 7, 2026

📊 Test Results for Test Example Applications

116b247 (2026_03_07_19_28_33)

IRONCLAD

Tested on 2026_03_07_19_28_33 at commit 116b247.

Test Checks TTFT (mean)TPS (mean)
llama_3.2_1b_prompt_13_tokens_1 ❌ 0/1 n/a n/a
llama_3.2_1b_prompt_13_tokens_40 ❌ 0/1 n/a n/a
llama_3.2_1b_prompt_2048_tokens_1 ❌ 0/1 n/a n/a
llama_3.2_1b_prompt_2048_tokens_40 ❌ 0/1 n/a n/a
📈 Trends (vs main branch) for Test Example Applications

116b247 (2026_03_07_19_28_33)

IRONCLAD Trends

llama_3.2_1b

Commit/Date Num Tokens (max)Num Tokens (mean)Num Tokens (median)Num Tokens (min)Num Tokens (stddev)TPS (max)TPS (mean)TPS (median)TPS (min)TPS (stddev)TTFT (max)TTFT (mean)TTFT (median)TTFT (min)TTFT (stddev)Total (max)Total (mean)Total (median)Total (min)Total (stddev)
130b6ea — 2025-12-05 21:33:1240.00 (+0.00%)40.00 (+0.00%)40.00 (+0.00%)40.00 (+0.00%)0.00 (n/a)4.71 (-0.42%)4.64 (-0.09%)4.64 (+0.65%)4.55 (-0.22%)0.05 (-17.66%)4.41 (-0.34%)4.39 (-0.19%)4.38 (-0.33%)4.37 (-0.15%)0.01 (-25.90%)12.96 (-0.00%)12.80 (+0.07%)12.80 (-0.23%)12.67 (+0.44%)0.09 (-21.12%)
0a6c11c — 2025-12-03 23:35:1540.00 (n/a)40.00 (n/a)40.00 (n/a)40.00 (n/a)0.00 (n/a)4.73 (n/a)4.64 (n/a)4.61 (n/a)4.56 (n/a)0.06 (n/a)4.42 (n/a)4.40 (n/a)4.40 (n/a)4.37 (n/a)0.02 (n/a)12.96 (n/a)12.79 (n/a)12.83 (n/a)12.62 (n/a)0.12 (n/a)

llama_3.2_1b_prompt_13_tokens_1

Commit/Date Num_Tokens (max)Num_Tokens (mean)Num_Tokens (median)Num_Tokens (min)Num_Tokens (stddev)TPS (max)TPS (mean)TPS (median)TPS (min)TPS (stddev)TTFT (max)TTFT (mean)TTFT (median)TTFT (min)TTFT (stddev)
116b247 — 2026-03-07 19:27:55n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)
897d04e — 2026-03-06 22:56:071.00 (n/a)1.00 (n/a)1.00 (n/a)1.00 (n/a)0.00 (n/a)0.00 (n/a)0.00 (n/a)0.00 (n/a)0.00 (n/a)0.00 (n/a)0.62 (n/a)0.62 (n/a)0.62 (n/a)0.62 (n/a)0.00 (n/a)

llama_3.2_1b_prompt_13_tokens_40

Commit/Date Num_Tokens (max)Num_Tokens (mean)Num_Tokens (median)Num_Tokens (min)Num_Tokens (stddev)TPS (max)TPS (mean)TPS (median)TPS (min)TPS (stddev)TTFT (max)TTFT (mean)TTFT (median)TTFT (min)TTFT (stddev)
116b247 — 2026-03-07 19:27:55n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)
897d04e — 2026-03-06 22:56:0740.00 (n/a)40.00 (n/a)40.00 (n/a)40.00 (n/a)0.00 (n/a)4.30 (n/a)4.30 (n/a)4.30 (n/a)4.30 (n/a)0.00 (n/a)0.61 (n/a)0.61 (n/a)0.61 (n/a)0.61 (n/a)0.00 (n/a)

llama_3.2_1b_prompt_2048_tokens_1

Commit/Date Num_Tokens (max)Num_Tokens (mean)Num_Tokens (median)Num_Tokens (min)Num_Tokens (stddev)TPS (max)TPS (mean)TPS (median)TPS (min)TPS (stddev)TTFT (max)TTFT (mean)TTFT (median)TTFT (min)TTFT (stddev)
116b247 — 2026-03-07 19:27:55n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)
897d04e — 2026-03-06 22:56:071.00 (n/a)1.00 (n/a)1.00 (n/a)1.00 (n/a)0.00 (n/a)0.00 (n/a)0.00 (n/a)0.00 (n/a)0.00 (n/a)0.00 (n/a)2.68 (n/a)2.68 (n/a)2.68 (n/a)2.68 (n/a)0.00 (n/a)

llama_3.2_1b_prompt_2048_tokens_40

Commit/Date Num_Tokens (max)Num_Tokens (mean)Num_Tokens (median)Num_Tokens (min)Num_Tokens (stddev)TPS (max)TPS (mean)TPS (median)TPS (min)TPS (stddev)TTFT (max)TTFT (mean)TTFT (median)TTFT (min)TTFT (stddev)
116b247 — 2026-03-07 19:27:55n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)n/a (n/a)
897d04e — 2026-03-06 22:56:0740.00 (n/a)40.00 (n/a)40.00 (n/a)40.00 (n/a)0.00 (n/a)4.00 (n/a)4.00 (n/a)4.00 (n/a)4.00 (n/a)0.00 (n/a)2.70 (n/a)2.70 (n/a)2.70 (n/a)2.70 (n/a)0.00 (n/a)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant