Difference between revisions of "GPU Checkpointing"
Jump to navigation
Jump to search
m |
m |
||
| Line 1: | Line 1: | ||
| − | This page explains how CRIU handles GPU-accelerated workloads, what vendor components are needed (NVIDIA & AMD), and how to use them for processes and containers. | + | This page explains how CRIU handles GPU-accelerated workloads, what vendor components are needed (NVIDIA & AMD), and how to use them for processes and containers. For more information, check out the [https://arxiv.org/abs/2502.16631 CRIUgpu paper], NVIDIA's [https://developer.nvidia.com/blog/checkpointing-cuda-applications-with-criu/ blog post], and AMD's [https://lpc.events/event/11/contributions/891/ LPC] and [https://indico.freedesktop.org/event/1/contributions/18/ X.Org] conference talks. |
CRIU checkpoints Linux-kernel resources (e.g., memory, threads, files, sockets). GPU state such as device memory, contexts, and queues that lives outside normal process address space needs special handling, so CRIU relies on vendor-specific [[plugins]]. | CRIU checkpoints Linux-kernel resources (e.g., memory, threads, files, sockets). GPU state such as device memory, contexts, and queues that lives outside normal process address space needs special handling, so CRIU relies on vendor-specific [[plugins]]. | ||
Revision as of 17:40, 29 October 2025
This page explains how CRIU handles GPU-accelerated workloads, what vendor components are needed (NVIDIA & AMD), and how to use them for processes and containers. For more information, check out the CRIUgpu paper, NVIDIA's blog post, and AMD's LPC and X.Org conference talks.
CRIU checkpoints Linux-kernel resources (e.g., memory, threads, files, sockets). GPU state such as device memory, contexts, and queues that lives outside normal process address space needs special handling, so CRIU relies on vendor-specific plugins.
NVIDIA CUDA
The checkpointing functionality for CUDA is enabled through a command-line utility called cuda-checkpoint. This utility allows to transparently checkpoint and restore the GPU state of a running Linux process.