| Precision | Peak (TFLOPS) | Per GPU (TFLOPS) | TFLOPS/W |
|---|---|---|---|
| FP64 | 136 | 34 | 32.38 |
| FP32 | 268 | 67 | 63.81 |
| TF32 | 1,978 | 495 | 941.90 |
| FP16 | 3,958 | 990 | 1884.76 |
| BF16 | 3,958 | 990 | 1884.76 |
| FP8 | 7,916 | 1,979 | 3769.52 |
| INT8 | 7,916 | 1,979 | 3769.52 |
All figures are dense. Vendors commonly headline the sparse number, which is twice the dense one.
NVLink
HGX H100 4-GPU configuration for smaller deployments