TBNilles/SparkyUI
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Initial commit: SparkyUI - ComfyUI for DGX Spark (Blackwell GB10)

Browse Source 
Docker-based ComfyUI setup for NVIDIA DGX Spark ARM64 + sm_121:
- CUDA 13.0.2 base (required for compute_121 support)
- PyTorch 2.9.1+cu130 ARM64 wheels
- SageAttention compiled with TORCH_CUDA_ARCH_LIST="12.1"
- Triton/torch.compile disabled (no sm_121 support yet)
- ComfyUI-Manager auto-installed at runtime
- Configurable model/data paths via .env

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
This commit is contained in:
Evan Carmen
2026-01-03 20:13:46 -06:00
commit 1f5aeb5248
12 changed files with 391 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.dockerignore
+30
View File
@@ -0,0 +1,30 @@
# Runtime data (mounted as volumes)
custom_nodes/
output/
input/
workflows/
# Git
.git/
.gitignore
# Environment and secrets
.env
*.env.local
# Documentation (not needed in image)
*.md
CLAUDE.md
README.md
LICENSE
# IDE
.vscode/
.idea/
# Python cache
__pycache__/
*.pyc
# Prebuilt wheels (built separately)
wheels/
.env.example
+16
View File
@@ -0,0 +1,16 @@
# SparkyUI - ComfyUI for DGX Spark (Blackwell GB10)
# Copy this to .env and customize paths as needed
# Base path where your existing ComfyUI installation lives (for models)
COMFYUI_HOST_PATH=/path/to/your/ComfyUI
# Base path for SparkyUI data (custom_nodes, outputs, inputs, etc.)
SPARKYUI_DATA_PATH=/path/to/SparkyUI
# ComfyUI settings
COMFYUI_PORT=8188
COMFYUI_FLAGS=--listen 0.0.0.0 --port 8188 --gpu-only
# Build refs (pin to specific commits/tags for reproducibility)
COMFYUI_REF=master
SAGEATTN_REF=main
.gitignore
+60
View File
@@ -0,0 +1,60 @@
# Project-specific internal docs
CLAUDE.md
# Environment (contains local paths)
.env
*.env.local
# Python
__pycache__/
*.py[cod]
*$py.class
*.so
.Python
build/
develop-eggs/
dist/
downloads/
eggs/
.eggs/
lib/
lib64/
parts/
sdist/
var/
*.egg-info/
.installed.cfg
*.egg
.venv/
venv/
# Docker
.docker/
# OS
.DS_Store
Thumbs.db
# IDE
.vscode/
.idea/
*.swp
*.swo
# Runtime directories - ignore contents but keep .gitkeep
custom_nodes/*
!custom_nodes/.gitkeep
output/*
!output/.gitkeep
input/*
!input/.gitkeep
workflows/*
!workflows/.gitkeep
# Wheels directory - for prebuilt ARM64/sm_121 binaries
# Ignore contents except .gitkeep (add wheels explicitly if needed)
wheels/*
!wheels/.gitkeep
Dockerfile
+58
View File
@@ -0,0 +1,58 @@
# CUDA 13.0 for Blackwell GB10 (sm_121 / compute_121)
# CUDA 12.8 only supports up to sm_120, but GB10 is sm_121.
# "devel" includes nvcc so we can compile CUDA extensions like SageAttention.
FROM nvidia/cuda:13.0.2-devel-ubuntu24.04
ARG DEBIAN_FRONTEND=noninteractive
ARG COMFYUI_REF=master
ARG SAGEATTN_REF=main
# Base system deps
RUN apt-get update && apt-get install -y --no-install-recommends \
git curl ca-certificates \
python3 python3-pip python3-venv python3-dev \
build-essential ninja-build cmake pkg-config \
&& rm -rf /var/lib/apt/lists/*
# Create venv (keeps python deps isolated inside container)
ENV VENV=/opt/venv
RUN python3 -m venv $VENV
ENV PATH="$VENV/bin:$PATH"
# Upgrade packaging tools
RUN pip install -U pip setuptools wheel
# ---- PyTorch (ARM64 + CUDA 13.0) ----
# PyTorch cu130 wheels work with CUDA 13.0.x runtime.
# ARM64 wheels available: torch-2.9.1+cu130, torchvision-0.24.1
RUN pip install --index-url https://download.pytorch.org/whl/cu130 \
torch torchvision
# ---- ComfyUI ----
RUN git clone https://github.com/comfyanonymous/ComfyUI.git /opt/ComfyUI && \
cd /opt/ComfyUI && \
git checkout ${COMFYUI_REF} || true
RUN pip install -r /opt/ComfyUI/requirements.txt
# ---- ComfyUI-Manager ----
# Handled at runtime by entrypoint.sh (clones if missing in mounted volume)
# This ensures latest version on each container start
# ---- SageAttention ----
# GB10 is compute capability 12.1 (sm_121).
# CUDA 13.0 NVCC supports sm_121, so we compile directly for it.
ENV TORCH_CUDA_ARCH_LIST="12.1"
ENV CUDA_HOME=/usr/local/cuda
# Build/install SageAttention from repo with sm_121 support
RUN pip install --no-build-isolation "git+https://github.com/thu-ml/SageAttention@${SAGEATTN_REF}" || true
# Expose ComfyUI
EXPOSE 8188
# Entry script handles runtime updates / flags
COPY entrypoint.sh /entrypoint.sh
RUN chmod +x /entrypoint.sh
ENTRYPOINT ["/entrypoint.sh"]
README.md
+144
View File
@@ -0,0 +1,144 @@
# SparkyUI
**ComfyUI + SageAttention for NVIDIA DGX Spark (Blackwell GB10)**
A Docker-based ComfyUI setup specifically engineered for the DGX Spark's unique ARM64 + Blackwell architecture.
## Why This Exists
The NVIDIA DGX Spark uses the **GB10 GPU** with compute capability **12.1 (sm_121)** - Blackwell architecture. This creates challenges:
| CUDA Version | Max Compute Capability | Can compile for GB10? |
|--------------|------------------------|----------------------|
| CUDA 12.8 | sm_120 | **No** |
| CUDA 13.0+ | sm_121 | **Yes** |
Standard ComfyUI containers and PyTorch wheels don't support sm_121. SparkyUI solves this by:
1. Using **CUDA 13.0.2** base image (supports sm_121)
2. Installing **PyTorch cu130** ARM64 wheels
3. Compiling **SageAttention** with `TORCH_CUDA_ARCH_LIST="12.1"`
4. Disabling **Triton/torch.compile** (doesn't support sm_121 yet)
## Quick Start
```bash
# Clone
git clone https://github.com/YOUR_USERNAME/SparkyUI.git
cd SparkyUI
# Configure paths
cp .env.example .env
# Edit .env with your paths
# Build (compiles SageAttention for sm_121 - takes ~10 min)
docker compose build
# Start
docker compose up -d
# View logs
docker compose logs -f
```
**Access:** http://localhost:8188 (or your DGX Spark's IP on LAN)
## Requirements
- **NVIDIA DGX Spark** (or other GB10-based system)
- **Docker** with NVIDIA Container Toolkit
- **NVIDIA Driver** 560+ (tested with 580.95)
- **~15GB** disk for Docker image
- **Models** from existing ComfyUI install (mounted read-only)
## Configuration
Copy `.env.example` to `.env` and edit:
```bash
# Path to your existing ComfyUI models (mounted read-only)
COMFYUI_HOST_PATH=/path/to/your/ComfyUI
# Path for SparkyUI data (custom_nodes, outputs, inputs)
SPARKYUI_DATA_PATH=/path/to/SparkyUI
# Optional: pin to specific versions
COMFYUI_REF=master
SAGEATTN_REF=main
```
## Architecture
```
┌─────────────────────────────────────────────────────────────┐
│ DGX Spark Host │
│ Ubuntu 24.04 (DGX OS 7) / Driver 580.x │
│ │
│ ┌─────────────────────────────────────────────────────┐ │
│ │ Docker Container (sparkyui:cu130) │ │
│ │ │ │
│ │ CUDA 13.0.2 + PyTorch 2.9.1+cu130 │ │
│ │ SageAttention 2.2.0 (compiled for sm_121) │ │
│ │ ComfyUI 0.7.x + ComfyUI-Manager │ │
│ │ │ │
│ │ Key env vars: │ │
│ │ TORCH_CUDA_ARCH_LIST="12.1" │ │
│ │ TORCHDYNAMO_DISABLE="1" │ │
│ └─────────────────────────────────────────────────────┘ │
│ │ │
│ Port 8188 (LAN) │
└─────────────────────────────────────────────────────────────┘
```
## Version Compatibility
Tested combinations:
| Component | Version | Notes |
|-----------|---------|-------|
| CUDA Base | 13.0.2 | Required for sm_121 |
| PyTorch | 2.9.1+cu130 | ARM64 wheel from PyTorch index |
| torchvision | 0.24.1+cu130 | ARM64 wheel |
| SageAttention | 2.2.0 | Compiled with sm_121 |
| ComfyUI | 0.7.0 | master branch |
| Driver | 580.95 | DGX OS 7 default |
## Known Limitations
1. **PyTorch Warning**: You'll see a warning about compute capability 12.1 being "outside supported range (8.0-12.0)". This is harmless - PyTorch works, and SageAttention's custom kernels are compiled natively.
2. **torch.compile Disabled**: Triton doesn't support sm_121 yet. `torch.compile()` is disabled via environment variables. Some nodes may run slower than on supported architectures.
3. **No GitHub Actions CI**: Can't build for ARM64 + sm_121 in GitHub's hosted runners. Must build locally on DGX Spark.
## Troubleshooting
### "no kernel image is available for execution on the device"
Your SageAttention wasn't compiled for sm_121. Rebuild:
```bash
docker compose build --no-cache
```
### PyTorch can't find CUDA
Ensure NVIDIA Container Toolkit is installed:
```bash
nvidia-ctk --version
docker run --rm --gpus all nvidia/cuda:13.0.2-base-ubuntu24.04 nvidia-smi
```
### ComfyUI-Manager missing
The entrypoint auto-clones it. Check logs:
```bash
docker compose logs | grep -i manager
```
## Future
When these land, SparkyUI can be simplified:
- [ ] PyTorch native sm_121 support → remove explicit `TORCH_CUDA_ARCH_LIST`
- [ ] Triton sm_121 support → remove `TORCHDYNAMO_DISABLE`
- [ ] SageAttention prebuilt ARM64 wheels → remove source build
## License
MIT
custom_nodes/.gitkeep
View File
docker-compose.yml
+53
View File
@@ -0,0 +1,53 @@
services:
comfyui:
build:
context: .
dockerfile: Dockerfile
args:
# Pin ComfyUI to a known-good commit/tag if desired
COMFYUI_REF: "${COMFYUI_REF:-master}"
# SageAttention ref (e.g., "main", "v2.2.0", or specific commit)
SAGEATTN_REF: "${SAGEATTN_REF:-main}"
image: sparkyui:cu130
container_name: comfyui
# GPU enablement
deploy:
resources:
reservations:
devices:
- driver: nvidia
count: all
capabilities: [gpu]
# LAN exposure
ports:
- "${COMFYUI_PORT:-8188}:8188"
environment:
COMFYUI_PORT: "${COMFYUI_PORT:-8188}"
COMFYUI_FLAGS: "${COMFYUI_FLAGS:---listen 0.0.0.0 --port 8188 --gpu-only}"
NVIDIA_VISIBLE_DEVICES: "all"
NVIDIA_DRIVER_CAPABILITIES: "compute,utility"
# Disable torch.compile/inductor - Triton doesn't support Blackwell sm_121a yet
TORCH_COMPILE_DISABLE: "1"
TORCHDYNAMO_DISABLE: "1"
volumes:
# Models from existing ComfyUI install (read-only)
- ${COMFYUI_HOST_PATH}/models:/opt/ComfyUI/models:ro
# Custom nodes - comment out to use container-only (fresh) custom_nodes
# If mounted, ComfyUI-Manager installs persist across container restarts
- ${SPARKYUI_DATA_PATH}/custom_nodes:/opt/ComfyUI/custom_nodes
# Outputs/inputs/workflows - persistent across restarts
- ${SPARKYUI_DATA_PATH}/output:/opt/ComfyUI/output
- ${SPARKYUI_DATA_PATH}/input:/opt/ComfyUI/input
- ${SPARKYUI_DATA_PATH}/workflows:/opt/ComfyUI/workflows
# Wheel cache (optional - for prebuilt wheels)
- ${SPARKYUI_DATA_PATH}/wheels:/opt/wheels
restart: unless-stopped
entrypoint.sh
+30
View File
@@ -0,0 +1,30 @@
#!/usr/bin/env bash
set -euo pipefail
COMFY_DIR="/opt/ComfyUI"
PORT="${COMFYUI_PORT:-8188}"
FLAGS="${COMFYUI_FLAGS:---listen 0.0.0.0 --port ${PORT}}"
echo "[entrypoint] Python: $(python --version)"
echo "[entrypoint] Torch: $(python -c 'import torch; print(torch.__version__)')"
echo "[entrypoint] CUDA: $(python -c 'import torch; print(torch.version.cuda)')"
echo "[entrypoint] Flags: ${FLAGS}"
# Ensure ComfyUI-Manager exists in mounted custom_nodes
# Check for __init__.py to detect corrupted/partial installs
if [[ ! -f "${COMFY_DIR}/custom_nodes/ComfyUI-Manager/__init__.py" ]]; then
echo "[entrypoint] ComfyUI-Manager missing or corrupted, cloning latest..."
rm -rf "${COMFY_DIR}/custom_nodes/ComfyUI-Manager" 2>/dev/null || true
git clone https://github.com/ltdrdata/ComfyUI-Manager.git \
"${COMFY_DIR}/custom_nodes/ComfyUI-Manager" || true
fi
# Install any requirements from custom nodes
for req in "${COMFY_DIR}"/custom_nodes/*/requirements.txt; do
if [[ -f "$req" ]]; then
echo "[entrypoint] Installing deps from: $req"
pip install -q -r "$req" || true
fi
done
exec python "${COMFY_DIR}/main.py" ${FLAGS}
input/.gitkeep
View File
output/.gitkeep
View File
wheels/.gitkeep
View File
workflows/.gitkeep
View File