Skip to content
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.

Sign up for GitHub

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

Jump to bottom

cuda : do not use batched GEMM when tensor cores are not available #3882

Merged
merged 1 commit into from
Nov 2, 2023
Merged

cuda : do not use batched GEMM when tensor cores are not available #3882

merged 1 commit into from
Nov 2, 2023

Conversation

ggerganov
Copy link
Owner

fix #3869

@young-developer young-developer mentioned this pull request Nov 1, 2023
Closed
@askmyteapot
Copy link

Can confirm the fix works for Pascal SM6.1

@ggerganov ggerganov added performance Speed related topics Nvidia GPU Issues specific to Nvidia GPUs labels Nov 1, 2023
@cebtenzzre
Copy link
Collaborator

cebtenzzre commented Nov 2, 2023 •
edited
Loading

I can confirm that this brings pp512 on my Tesla P40 back to pre-#3749 speeds.

Now both #3749 and #3776 can be worked around via -DLLAMA_CUDA_FORCE_MMQ=ON on older cards.

@ggerganov ggerganov merged commit 4d719a6 into master Nov 2, 2023
33 checks passed
@ggerganov ggerganov deleted the try-fix-3869 branch November 2, 2023 06:35
@Ph0rk0z Ph0rk0z mentioned this pull request Nov 5, 2023
Closed
olexiyb pushed a commit to Sanctum-AI/llama.cpp that referenced this pull request Nov 23, 2023
@ggerganov @olexiyb
cuda : check if this fixes Pascal card regression (ggerganov#3882)
a660fc3
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@cebtenzzre cebtenzzre cebtenzzre approved these changes

@young-developer young-developer young-developer approved these changes

Assignees
No one assigned
Labels
Nvidia GPU Issues specific to Nvidia GPUs performance Speed related topics
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

CTX Processing regression for Pascal - Commit 2b4ea35
4 participants
@ggerganov @askmyteapot @cebtenzzre @young-developer