CUDA: use 1 thread if model is fully offloaded#2915
Merged
ggerganov merged 1 commit intoggml-org:masterfrom Sep 21, 2023
Merged
CUDA: use 1 thread if model is fully offloaded#2915ggerganov merged 1 commit intoggml-org:masterfrom
ggerganov merged 1 commit intoggml-org:masterfrom
Conversation
Contributor
|
Maybe update the help for It might be good to mention the ability to set it to |
|
Using 1 vs 15 for me does make it .2-.5 tokens faster. |
ggerganov
requested changes
Sep 2, 2023
Member
ggerganov
left a comment
ggerganov
left a comment
There was a problem hiding this comment.
Better to implement this entirely in llama_eval_internal, similar to what has been done for BLAS:
Comment on lines
731
to
735
Member
There was a problem hiding this comment.
This is an implementation detail that the user should not need to know and in the future we will fix this anyway
Contributor
Author
|
I haven't forgotten about this, I've only prioritized other things because I think that this is not that high priority. |
JohannesGaessler
force-pushed
the
cuda-n-threads-fix
branch
from
ff65f9a to
de8035a
Compare
Contributor
Author
|
@ggerganov thank you for the hint with |
slaren
approved these changes
Sep 18, 2023
ggerganov
approved these changes
Sep 21, 2023
pkrmf
pushed a commit
to morlockstudios-com/llama.cpp
that referenced
this pull request
Sep 26, 2023
Seunghhon
pushed a commit
to Seunghhon/llama.cpp
that referenced
this pull request
Apr 26, 2026
phuongncn
pushed a commit
to phuongncn/llama.cpp-gx10-dgx-sparks-deepseekv4
that referenced
this pull request
Apr 28, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Currently when using the maximum possible number of GPU layers with CUDA there is no benefit from > 1 thread. In fact, using more than 1 thread is detrimental due to increased overhead. This PR changes the logic for the default number of threads in such a way that (unless the user manually overrides it) only a single thread is used if all layers are offloaded.
I also changed the logic for
llama-benchto be the same asmain: -1 is interpreted as the number of logical cores.