Hardware support
Use this page to check whether a prebuilt binary exists for your accelerator. If you are installing on a normal workstation or server, start with the quickstart and let the installer choose.
Accelerators
Section titled “Accelerators”| Platform | Acceleration | Prebuilt binary |
|---|---|---|
| Linux x86_64 + NVIDIA GPU | CUDA (Ampere and newer) | yes, per compute capability and driver CUDA support |
| Linux aarch64 + NVIDIA GPU | CUDA (Grace: GH200/GB200/GB10) | yes, sm90/100/121, per driver CUDA support |
| Apple Silicon (macOS arm64) | Metal | yes |
| Linux x86_64 / aarch64, no GPU | CPU | yes |
| Windows x86_64 | CPU | yes |
| Intel Mac, unlisted GPU | source build | no |
NVIDIA compute capabilities
Section titled “NVIDIA compute capabilities”The minimum supported NVIDIA GPU is Ampere (compute capability 8.0). Turing (sm75: RTX 20-series, GTX 16-series, Tesla T4) and older are not supported by current prebuilts.
| Compute capability | Architecture | Representative GPUs |
|---|---|---|
| 8.0 | Ampere (datacenter) | A100, A30 |
| 8.6 | Ampere (consumer) | RTX 3090/3080/3070/3060, A40, A10 |
| 8.9 | Ada | RTX 4090/4080, L40, L4 |
| 9.0 | Hopper | H100, H200 |
| 10.0 | Blackwell (datacenter) | B200, GB200 |
| 12.0 | Blackwell (consumer) | RTX 5090/5080 |
| 12.1 | GB10 | DGX Spark |
CUDA artifacts
Section titled “CUDA artifacts”CUDA artifact names encode both the toolkit lane and compute capability:
mistralrs-cuda128-sm90-aarch64-unknown-linux-gnu.tar.gzThe installer chooses the newest published lane that the installed NVIDIA driver can load.
| Driver reports | Artifact lane | Notes |
|---|---|---|
CUDA 13.3+ on Hopper / sm90 | cuda133 | cuTile lane for Hopper |
CUDA 13.2+ on Ampere/Ada / sm80, sm86, sm89 | cuda132 | cuTile lane for Ampere and Ada |
CUDA 13.1+ on Blackwell / sm100, sm120, sm121 | cuda131 | cuTile lane for Blackwell and GB10 |
| CUDA 13.0+ | cuda130 | CUDA 13 baseline lane |
CUDA 12.9+ on GB10 / sm121 | cuda129 | needed because CUDA 12.8 does not target sm121 |
| CUDA 12.8+ | cuda128 | baseline lane for Ampere and newer |
| Architecture | Compute capabilities |
|---|---|
| x86_64 | 80, 86, 89, 90, 100, 120 |
| aarch64 (NVIDIA Grace: GH200, GB200, GB10/DGX Spark) | 90, 100, 121 |
| Asset token | Built with | Minimum nvidia-smi CUDA version | Published compute capabilities | cuTile |
|---|---|---|---|---|
cuda128 | CUDA 12.8.1 | 12.8 | x86_64: 80, 86, 89, 90, 100, 120; aarch64: 90, 100 | no |
cuda129 | CUDA 12.9.1 | 12.9 | aarch64: 121 | no |
cuda130 | CUDA 13.0.0 | 13.0 | x86_64: 80, 86, 89, 90, 100, 120; aarch64: 90, 100, 121 | no |
cuda131 | CUDA 13.1.2 | 13.1 | x86_64: 80, 86, 89, 90, 100, 120; aarch64: 90, 100, 121 | sm100, sm120, sm121 |
cuda132 | CUDA 13.2.0 | 13.2 | x86_64: 80, 86, 89 | sm80, sm86, sm89 |
cuda133 | CUDA 13.3.0 | 13.3 | x86_64: 90; aarch64: 90 | sm90 |
Each artifact bundles the CUDA runtime libraries it needs, so no CUDA toolkit is required at runtime. The installed NVIDIA driver still has to be new enough for that artifact’s toolkit lane.
The same compatibility lanes are used by the Docker images and CUDA Python wheels. See Python getting started for wheel install commands.
Feature availability by architecture
Section titled “Feature availability by architecture”| Feature | Requirement |
|---|---|
flash-attn (v2) | compute capability 8.0+ |
flash-attn-v3 | Hopper (9.0) |
| FP8 matmul | compute capability 8.9+ |
| cuTile MoE backend | Ampere/Ada (8.x) with CUDA >= 13.2; Hopper (9.0) with CUDA >= 13.3; Blackwell+ (10.x/12.x) with CUDA >= 13.1 |
| CUTLASS MoE backend | compute capability 8.0+ |
See cargo features for the feature flags and MoE expert backends for backend selection.