Skip to content

Perf Tracker 2026 11 13

github-actions[bot] edited this page Nov 15, 2026 · 1 revision

Perf Tracker 2026-11-13

Nightly run 43264767627, commit 9bf4bf56c, NVIDIA H200, CUDA 13.2, driver 580.178.04, torch 2.13.0+cu132, TileLang 0.1.12, image ghcr.io/tile-ai/tileops-ci-runner:local-runner-2.338.2. Times are CUPTI device_busy_ms.

A row is behind when its fastest comparator's time over TileOPs' time is below 0.99. A row whose fastest comparator takes under 10 us is behind only if it was also below that ratio the night before. An op is labelled by its worst row: P0 < 0.50, P1 < 0.80, P2 < 0.90, P3 < 0.95, P4 < 0.99.

Summary

Measure Value
Ops behind 7 of 206
Rows behind 8 of 2337
By label P0 0, P1 0, P2 0, P3 0, P4 7
Failed rows 0

Since nightly run 43063959201:

  • Newly behind: ExpFwdOp, FloorFwdOp, RoundFwdOp, TruncFwdOp
  • Caught up: AddFwdOp, GEMMW4A16FwdOp
  • Not measured tonight: none

Queue

Nightlies behind counts the consecutive nightlies, up to 30, in which the op had a row behind.

# Op Category Label Worst ratio Rows behind Nightlies behind
1 SoftmaxFwdOp reduction P4 0.9757 2 of 42 10
2 GroupedGEMMFwdOp gemm P4 0.9774 1 of 10 9
3 ErfFwdOp elementwise P4 0.9820 1 of 5 3
4 TruncFwdOp elementwise P4 0.9886 1 of 5 1
5 RoundFwdOp elementwise P4 0.9896 1 of 5 1
6 FloorFwdOp elementwise P4 0.9896 1 of 5 1
7 ExpFwdOp elementwise P4 0.9896 1 of 5 1

Rows behind

SoftmaxFwdOp

Benchmark benchmarks.ops.bench_softmax.

Workload TileOPs (us) Fastest comparator Comparator (us) Ratio
quack-throughput-32768-float16 1051.020 quack 1025.430 0.9757
quack-throughput-32768-bfloat16 1077.450 quack 1063.540 0.9871

GroupedGEMMFwdOp

Benchmark benchmarks.ops.bench_grouped_gemm.

Workload TileOPs (us) Fastest comparator Comparator (us) Ratio
qwen3-30b-ep8-skew-down-bfloat16 42.432 quack 41.473 0.9774

ErfFwdOp

Benchmark benchmarks.ops.bench_elementwise.

Workload TileOPs (us) Fastest comparator Comparator (us) Ratio
16M-float32 32.001 torch 31.425 0.9820

TruncFwdOp

Benchmark benchmarks.ops.bench_elementwise.

Workload TileOPs (us) Fastest comparator Comparator (us) Ratio
16M-float32 30.816 torch-compile 30.464 0.9886

RoundFwdOp

Benchmark benchmarks.ops.bench_elementwise.

Workload TileOPs (us) Fastest comparator Comparator (us) Ratio
16M-float32 30.817 torch-compile 30.496 0.9896

FloorFwdOp

Benchmark benchmarks.ops.bench_elementwise.

Workload TileOPs (us) Fastest comparator Comparator (us) Ratio
16M-float32 30.785 torch-compile 30.465 0.9896

ExpFwdOp

Benchmark benchmarks.ops.bench_elementwise.

Workload TileOPs (us) Fastest comparator Comparator (us) Ratio
16M-float32 30.849 torch-compile 30.529 0.9896

Clone this wiki locally