intel-device-plugins-for-ku.../cmd/gpu_levelzero
Mikko Ylinen 3e141cc736 ci: move to golangci-lint v1.63.1
along with it, fix some wsl findings.

Signed-off-by: Mikko Ylinen <mikko.ylinen@intel.com>
2025-01-02 12:00:34 +02:00
..
main_test.go gpu: add levelzero application 2024-09-19 19:14:15 +03:00
main.go ci: move to golangci-lint v1.63.1 2025-01-02 12:00:34 +02:00
README.md gpu: levelzero: documentation 2024-09-19 19:14:15 +03:00
ze.c gpu: add levelzero application 2024-09-19 19:14:15 +03:00
ze.h levelzero: use define for error temperature 2024-09-23 08:27:45 +03:00
zes.c levelzero: use define for error temperature 2024-09-23 08:27:45 +03:00

Intel GPU Level-Zero sidecar

Table of Contents

Introduction

Intel GPU Level-Zero sidecar is an extension for the Intel GPU plugin to query additional GPU details from the oneAPI/Level-Zero API. As the Level-Zero is a C/C++ API, it is preferred to keep the original GPU plugin as-is and add the additional functionality via the Level-Zero sidecar. The GPU plugin can be configured to use the Level-Zero sidecar with an overlay, see install.

Intel GPU plugin and the Level-Zero sidecar communicate via gRPC on a local socket visible only to the containers.

Note

: Intel Device Plugin Operator doesn't yet support enabling Level-Zero sidecar in the GPU CR object.

Modes and Configuration Options

Flag Argument Default Meaning
-socket unix socket path /var/lib/levelzero/server.sock Unix socket path which the server registers itself into.
-wsl - disabled Adapt sidecar to run in the WSL environment.
-v verbosity 1 Log verbosity

Install

Installing the sidecar along with the GPU plugin happens via two possible overlays: health and wsl.

Health overlay adds the sidecar to the base GPU plugin deployment and configures GPU plugin to retrieve device health indicators from the Level-Zero API:

$ kubectl -k deployments/gpu_plugin/overlays/health

WSL layer enables Intel GPU detection with WSL (Windows Subsystem for Linux) Kubernetes clusters. It also leverages the Level-Zero sidecar:

$ kubectl -k deployments/gpu_plugin/overlays/wsl