Compatibility
Built for the GPUs you already run.
GPU architectures, precisions, and runtimes supported by the Kernova AI optimization engine.
| GPU | Architecture | FP8 | BF16 | FP16 | INT8 |
|---|---|---|---|---|---|
| H100 / H200 (SXM, PCIe) | Hopper | Beta | Supported | Supported | Supported |
| A100 40GB / 80GB | Ampere | Planned | Supported | Supported | Supported |
| L40S, L4 | Ada Lovelace | Beta | Supported | Supported | Beta |
| RTX 4090 / 6000 Ada | Ada Lovelace | Planned | Beta | Supported | Beta |
| B200 | Blackwell | Planned | Planned | Planned | Planned |
Frameworks & runtimes
- PyTorch 2.1+Supported
- Triton 2.x / 3.xSupported
- CUDA 12.xSupported
- vLLMBeta
- Hugging Face TGIBeta
- TensorRT-LLMPlanned
Support matrix reflects the current roadmap and should be confirmed against your release notes before launch.