DGX Spark

FrontPage

GB10 †

forum †

https://forums.developer.nvidia.com/c/accelerated-computing/dgx-spark-gb10/719

DGX OS用 †

ユーザーガイド
https://docs.nvidia.com/dgx/dgx-spark/index.html

Start Building on DGX Spark
https://build.nvidia.com/spark

UEFI †

電源を入れてからEscかDelete連打で入れる
PXE bootが先頭になっていたので起動順を入れ替え。

firmware更新 †

fwupdmgr get-upgrades

docker †

/etc/docker/daemon.json
nvidia runtimeを標準で有効化

{
    "default-runtime": "nvidia",
    "runtimes": {
        "nvidia": {
            "path": "nvidia-container-runtime",
            "runtimeArgs": []
        }
    }
}

ユーザーからdockerを使用

sudo gpasswd -a $USER docker

確認

docker run --rm --gpus all nvidia/cuda:12.6.0-base-ubuntu24.04 nvidia-smi

docker上でのcuda関係
https://docs.nvidia.com/dgx/dgx-spark/nvidia-container-runtime-for-docker.html
確認

docker run -it --runtime=nvidia --gpus=all nvcr.io/nvidia/cuda:13.0.1-devel- ubuntu24.04 bash
# Inside the container:
nvidia-smi
ls /dev/nvidia*

出力

/dev/nvidia-modeset  /dev/nvidia-uvm  /dev/nvidia-uvm-tools  /dev/nvidia0  /dev/nvidiactl

/dev/nvidia-caps:
nvidia-cap1  nvidia-cap2

cuda関係その他
https://docs.nvidia.com/dgx/dgx-spark/ngc.html

network †

avahi-daemonとsshdはデフォルトで動いているので$HOSTNAMEでアクセス可能

LM Studio †

Download

curl https://lmstudio.ai/download/latest/linux/arm64 -o LM-Studio-latest.AppImage
chmod +x LM-Studio-latest.AppImage

AppArmor?の設定に引っかかるのでsandboxを無効化しないと起動できない

./LM-Studio-latest.AppImage --no-sandbox

vLLM †

https://build.nvidia.com/spark/vllm/instructions
docker image
https://catalog.ngc.nvidia.com/orgs/nvidia/containers/vllm

docker pull nvcr.io/nvidia/vllm:25.12.post1-py3

別イメージ https://huggingface.co/nologik/vllm-dgx-spark

docker pull avarok/vllm-dgx-spark:v11
docker run --rm -it --gpus all --ipc=host -p 8000:8000 -e VLLM_FLASHINFER_MOE_BACKEND=latency -v ~/.cache/huggingface:/root/.cache/huggingface avarok/vllm-dgx-spark:v11 serve cybermotaz/nemotron3-nano-nvfp4-w4a16 --quantization modelopt_fp4 --kv-cache-dtype fp8 --trust-remote-code --max-model-len 131072 --gpu-memory-utilization 0.85

model
https://huggingface.co/Firworks/GLM-4.5-Air-nvfp4

起動

docker run -it --gpus all -p 8000:8000 --ipc=host --ulimit memlock=-1 --ulimit stack=67108864 nvcr.io/nvidia/vllm:25.12.post1-py3 vllm serve "Firworks/GLM-4.5-Air-nvfp4" --served-model-name glm45a --enable-auto-tool-choice --tool-call-parser glm45

LM Studio †

CLIのみでの接続制限解除
https://github.com/lmstudio-ai/lms/issues/201

~.lmstudio/.internal/http-server-config.json

claude code †

export ANTHROPIC_BASE_URL=http://hostname:8000
export ANTHROPIC_AUTH_TOKEN=none
claude --model glm4a5

ubuntuの設定 †

keyboard †

sudo vi /etc/default/keyboard
# KEYBOARD CONFIGURATION FILE

# Consult the keyboard(5) manual page.

XKBMODEL="pc105"
XKBLAYOUT="us"
XKBVARIANT=""
XKBOPTIONS="ctrl:nocaps"

BACKSPACE="guess"

フォルダ名変更 †

LANG=C xdg-user-dirs-gtk-update

tailscale †

curl -fsSL https://tailscale.com/install.sh | sh

samba †

sudo systemctl start smbd
sudo systemctl enable smbd
sudo vi /etc/samba/smb.conf
sudo smbpasswd -a $USER

日本語とデスクトップ変更 †

sudo apt install ibus-mozc language-pack-ja xubuntu-desktop

uv †

curl -LsSf https://astral.sh/uv/install.sh | sh