Skip to content

cuda: use the 4-column GDN warp layout on all Ampere+ NVIDIA, not only GB10 (+6% prefill on Ada) - #216

Merged
bri-prism merged 2 commits into
PrismML-Eng:prismfrom
professorpalmer:cuda-gdn-cols-ampere
Sep 23, 2026
Merged

cuda: use the 4-column GDN warp layout on all Ampere+ NVIDIA, not only GB10 (+6% prefill on Ada)#216
bri-prism merged 2 commits into
PrismML-Eng:prismfrom
professorpalmer:cuda-gdn-cols-ampere

cuda: keep only the GDN cols-per-warp launch invariant in comments.

937c4d8
Select commit
Loading
Failed to load commit list.
Sign in for the full log view

Annotations

1 warning
labeler
succeeded Sep 21, 2026 in 55s