Skip to content

[CUDAGraph]CUDA Graph support unique memory pool - #4230

Merged
Jiang-Jia-Jun merged 5 commits into
PaddlePaddle:developfrom
gongshaotian:unique_memory_pool
Sep 24, 2025
Merged

Jiang-Jia-Jun merged 5 commits into
PaddlePaddle:developfrom
gongshaotian:unique_memory_pool

Conversation

@gongshaotian

@gongshaotian gongshaotian commented Sep 23, 2025

Copy link
Copy Markdown
Collaborator

Using a unified memory pool during CUDA Graph capture can reduce the video memory overhead of CUDA Graph. In earlier versions, CUDA Graph used the unified memory pool to generate bugs. This risk has not been completely eliminated, so it is not enabled by default.

Related PR: #4199

@paddle-bot

paddle-bot Bot commented Sep 23, 2025

Copy link
Copy Markdown

Thanks for your contribution!

lizhenyun01
lizhenyun01 previously approved these changes Sep 24, 2025
Comment thread fastdeploy/model_executor/graph_optimization/cudagraph_piecewise_backend.py Outdated
Comment thread fastdeploy/model_executor/graph_optimization/cudagraph_piecewise_backend.py Outdated

@carryyu carryyu left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@Jiang-Jia-Jun
Jiang-Jia-Jun merged commit 870364b into PaddlePaddle:develop Sep 24, 2025
33 of 39 checks passed
@gongshaotian
gongshaotian deleted the unique_memory_pool branch November 3, 2025 06:49
xiaoguoguo626807 pushed a commit to xiaoguoguo626807/FastDeploy that referenced this pull request May 7, 2026
* cuda graph use unique memory pool

* fix custom device import bug

* refine code

* refine code

* refine code
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants