Hardware and runtimes
Choose your runtime first, then check the model and attention backend. A model being listed does not mean every precision, kernel, or device supports it.
| Hardware | Runtime | Start here | What changes |
|---|---|---|---|
| NVIDIA RTX / workstation GPUs | PyTorch + CUDA | NVIDIA GPU setup, then the model’s cookbook | Memory budget and GPU architecture determine which precision and attention kernels are available. The RTX PRO 6000 FastH3 guide is a specific Blackwell recipe, not a requirement for all NVIDIA users. |
| NVIDIA data center GPUs | PyTorch + CUDA | NVIDIA GPU setup, then multi-GPU configuration | Use the configuration and optimization guides for parallelism, offloading and backend selection. Keep benchmark results tied to the measured GPU. |
| NVIDIA DGX Spark / GB10 | PyTorch + CUDA on ARM64 | Spark installation | This is a separate installation path. Pairing two Sparks and tuning are advanced steps after one device works. |
| Apple Silicon | Native MLX | MLX installation | Use MLX-compatible checkpoints and examples. FastMetal and FastH3 have different conversion and generation paths. The older MPS example is a separate PyTorch demo. |
| AMD GPUs | PyTorch + ROCm | ROCm support and current limits | The repository has a ROCm platform and Triton attention routes. CUDA-specific kernels and quantized exports do not automatically apply. |
Then choose a model
Fast Models highlights FastVideo releases. The support matrix covers the wider model set. The cookbook provides model-specific recipes and runtime choices. Follow a recipe’s recorded requirements rather than inferring memory fit from a model name or parameter count.
Shared guides
- Configuration owns generation settings and parallelism.
- Offloading explains memory tradeoffs.
- Optimizations lists attention and precision constraints.
- ComfyUI is an integration, not a hardware platform.