metal : use residency sets #11427
Merged
metal : use residency sets #11427
+119
−17
Conversation
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
ggerganov
force-pushed
the
gg/metal-residency-sets
branch
from
febb813 to
4dad9fa
Compare
github-actions
bot
added
ggml
|
Great news - this change finally resolves the annoying overhead that I was observing. The only remaining question is how to implement this to be compatible with macOS < 15.0. Any suggestions? Edit: resolved |
ggml-ci
ggerganov
force-pushed
the
gg/metal-residency-sets
branch
from
21850f6 to
2674f02
Compare
Animaxx
added a commit
to Animaxx/llama.cpp
that referenced
this pull request
Jan 28, 2025
metal : use residency sets
tinglou
pushed a commit
to tinglou/llama.cpp
that referenced
this pull request
Feb 13, 2025
* metal : use residency sets ggml-ci * metal : restore commandBufferWithUnretainedReferences calls [no ci] * metal : release descriptors ggml-ci * metal : check env GGML_METAL_NO_RESIDENCY ggml-ci * metal : fix build + clean-up ggml-ci
arthw
pushed a commit
to arthw/llama.cpp
that referenced
this pull request
Feb 26, 2025
* metal : use residency sets ggml-ci * metal : restore commandBufferWithUnretainedReferences calls [no ci] * metal : release descriptors ggml-ci * metal : check env GGML_METAL_NO_RESIDENCY ggml-ci * metal : fix build + clean-up ggml-ci
mglambda
pushed a commit
to mglambda/llama.cpp
that referenced
this pull request
Mar 8, 2025
* metal : use residency sets ggml-ci * metal : restore commandBufferWithUnretainedReferences calls [no ci] * metal : release descriptors ggml-ci * metal : check env GGML_METAL_NO_RESIDENCY ggml-ci * metal : fix build + clean-up ggml-ci
|
@ggerganov is there any reason we wouldn't set GGML_METAL_NO_RESIDENCY=1 on macOS? |
|
Without residency sets you will hit the issue from #10119 - that was the main reason to introduce them. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
fix #10119
Using residency sets makes the allocated memory stay wired and eliminates almost completely the overhead observed in #10119. For example, on M2 Ultra, using 7B Q8_0 model the requests are ~250ms faster thanks to this change. It seems it is not necessary to attach the residency sets to the command queue and buffers, so the change is rather simple. For each buffer, we create an associated
MTLResidencySetand add theMTLBufferobjects to it. After that we commit it and request residency:https://github.com/ggerganov/llama.cpp/blob/225d2e0ca1d7a7e627f2cea4a43dd77a83b9f078/ggml/src/ggml-metal/ggml-metal.m#L1084-L1091
build: b9126fe (4561)
Metal backend changes
Checks the environment variable
GGML_METAL_NO_RESIDENCY. If set, then no residency sets will be created, allowing the GPU memory to be collected by the OS after 1 second of inactivity. Generally, this should rarely be needed as it hurts the performance of the application, but keeping support just in case.