Prerequisites

  • Architecture: x86_64
  • Chip models: Kunlunxin P800
  • Host driver: 5.37.1
  • Container toolkit (optional) : xpu_container >= 1.0.13

Image contents

Base image

ubuntu:22.04

System packages

Explicitly installed; the version is the one baked into this image:

  • build-essential — 12.9ubuntu3
  • ca-certificates — 20260601~22.04.1
  • cmake — 3.22.1
  • curl — 7.81.0
  • g++ — 11.2.0
  • gcc — 11.2.0
  • git — 2.34.1
  • make — 4.3
  • pciutils — 3.7.0
  • unzip — 6.0
  • vim — 8.2.3995

SDK components

  • CUDA 12.9.0_575.51.03
  • XRE-CUDA12 5.37.1.0
  • XCUDART 5.13.0

Environment

  • PATH=/usr/local/xpu/bin:$PATH
  • LD_LIBRARY_PATH=/usr/local/xpu/lib:/usr/local/xcudart/lib

Launch

With the container toolkit (optional):

  docker run --rm -it \
  --runtime xpu \
  -e CXPU_VISIBLE_DEVICES=0 \
  harbor.baai.ac.cn/flagos-base/flagos-base-kunlunxin-xre5.37.1:2.1.2 bash
  

Without a toolkit — plain docker / podman:

  docker run --rm -it \
  --device /dev/xpu0 \
  --device /dev/xpuctrl \
  harbor.baai.ac.cn/flagos-base/flagos-base-kunlunxin-xre5.37.1:2.1.2 bash
  

Verify

Inside the container, confirm the accelerator is visible:

  xpu-smi
  

Last updated: 2026-08-14 21:51:46 · 521f2968843d

Last updated 15 Aug 2026, 09:57 +0800. history