---
slug: getting_started/hardware
title: 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](/FastVideo/getting_started/installation/gpu), then the model's cookbook | Memory budget and GPU architecture determine which precision and attention kernels are available. The [RTX PRO 6000 FastH3 guide](/FastVideo/inference/fasth3_rtx_pro_6000) is a specific Blackwell recipe, not a requirement for all NVIDIA users. |
| NVIDIA data center GPUs | PyTorch + CUDA | [NVIDIA GPU setup](/FastVideo/getting_started/installation/gpu), then [multi-GPU configuration](/FastVideo/inference/configuration#multi-gpu-setup) | 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](/FastVideo/getting_started/installation/spark) | This is a separate installation path. [Pairing two Sparks](/FastVideo/getting_started/installation/spark_pair) and [tuning](/FastVideo/getting_started/installation/spark_performance) are advanced steps after one device works. |
| Apple Silicon | Native MLX | [MLX installation](/FastVideo/getting_started/installation/mlx) | 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](/FastVideo/getting_started/installation/amd) | 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](/FastVideo/fast-models/) highlights FastVideo releases.
The [support matrix](/FastVideo/inference/support_matrix) covers the wider model set.
The [cookbook](/FastVideo/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](/FastVideo/inference/configuration) owns generation settings and parallelism.
- [Offloading](/FastVideo/inference/offloading) explains memory tradeoffs.
- [Optimizations](/FastVideo/inference/optimizations) lists attention and precision constraints.
- [ComfyUI](/FastVideo/inference/comfyui) is an integration, not a hardware platform.