New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
feat(hardware support): verify if hopper optimizations apply to ada lovelace (sm_89) #2192
Comments
jon-chuang
changed the title
feat(hardware support): check if hopper optimizations apply to ada lovelace (sm_89)
feat(hardware support): verify if hopper optimizations apply to ada lovelace (sm_89)
Aug 28, 2023
|
Cool, let me see if TMA will improve sm_89 (Ada lovelace) performance or result in any errors. |
I found some evidence on
Use of TMA results in a Python abort e.g.
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Although server class GPU (A100, H100) are main target for production, some may have their own server with commercial GPU or want to develop optimizations on local GPU. Hence, we should attempt to support sm_89 for hopper-specific features. Currently they might be ignored.
triton/python/tutorials/09-experimental-tma-matrix-multiplication.py
Line 37 in 5f448b2
Actually, NVIDIA docs seem pretty clear:
Seems that the only thing that hopper and ada lovelace (or other sm_89) share is support for fp8 tensor core, which is not listed explicitly.
The text was updated successfully, but these errors were encountered: