Why does kernl not support gpus before ampere architecture? #276
Answered
by
jonathlela
seastar105
asked this question in
Q&A
-
|
as @jonathlela said in #246, kernl doesn't support gpus before ampere(8.0), but in triton's readme, triton says they support compute capability 7.0+. what's the reason kernl not supporting volta, turing architecture? |
Beta Was this translation helpful? Give feedback.
Answered by
jonathlela
Feb 8, 2023
Replies: 1 comment
-
|
Currently, our attention kernels don't compile with triton on turing/volta, even the latest version. We didn't investigate much but it may be related to triton-lang/triton#616. We may be able to rewrite the forward pass of the kernel to make it works but we prefer to wait for a more stable release of triton before. |
Beta Was this translation helpful? Give feedback.
0 replies
Answer selected by
seastar105
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Currently, our attention kernels don't compile with triton on turing/volta, even the latest version. We didn't investigate much but it may be related to triton-lang/triton#616. We may be able to rewrite the forward pass of the kernel to make it works but we prefer to wait for a more stable release of triton before.