Skip to content

smoke: fix CUDA device assert and tg128 column#42

Merged
worthant merged 1 commit into
devfrom
fix/smoke-cuda-assert
Jul 22, 2026
Merged

smoke: fix CUDA device assert and tg128 column#42
worthant merged 1 commit into
devfrom
fix/smoke-cuda-assert

Conversation

@worthant

Copy link
Copy Markdown
Member

Manual validation run of the committed smoke scripts on a rented RTX 5090 (dev-latest, commit f2757df):

  • linux-x64-vulkan: SMOKE OK — pp512 25.4k t/s, tg128 674 t/s
  • linux-x64-cuda-13.3: worked perfectly (CUDA backend loaded, pp512 60.5k t/s, tg128 1349 t/s — 2.2–2.4× the Vulkan path on the same GPU) — but was reported FAIL by a bug in my assert: the regex required cuda and NVIDIA on the same log line, while the bench log prints the device as Device 0: NVIDIA GeForce RTX 5090, compute capability 12.0.

Fixes:

  • CUDA device assert → Device [0-9]+: NVIDIA (validated against the real bench.log from the failing run)
  • tg128 extraction read the test-name column ($7) instead of the t/s column ($8)

The CUDA device line in the bench log is 'Device 0: NVIDIA ...' with no
'cuda' on the same line, so the combined regex never matched and a fully
working CUDA run (60k t/s pp512 on an RTX 5090) was reported as FAIL.
Also the tg128 extraction read the test-name column instead of the t/s
column. Both asserts validated against the real bench.log from that run.
@worthant
worthant merged commit 42c197b into dev Jul 22, 2026
11 checks passed
@worthant
worthant deleted the fix/smoke-cuda-assert branch July 22, 2026 03:06
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant