Changes

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 conference talks at  [https://lpc.events/event/11/contributions/891/ LPC] and [https://indico.freedesktop.org/event/1/contributions/18/ X.Org].

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 conference talks at  [https://lpc.events/event/11/contributions/891/ LPC] and [https://indico.freedesktop.org/event/1/contributions/18/ X.Org].

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]].

== How CRIU integrates with GPU checkpointing mechanisms? ==

The checkpointing functionality for CUDA is enabled through a command-line utility called [http://github.com/NVIDIA/cuda-checkpoint cuda-checkpoint]. This utility allows to transparently checkpoint and restore the GPU state of a running Linux process.

=== CUDA Plugin ===

== AMD ROCm ==

 

=== AMDGPU Plugin ===