| Precision | Peak (PFLOPS) | Per GPU (TFLOPS) | TFLOPS/W |
|---|---|---|---|
| FP64 | 0.2 | 30.0 | — |
| FP32 | 0.5 | 60.0 | — |
| TF32 | 7.0 | 875.0 | — |
| FP16 | 14.0 | 1750.0 | — |
| BF16 | 14.0 | 1750.0 | — |
| FP8 | 28.0 | 3500.0 | — |
| FP4 | 56.0 | 7000.0 | — |
| INT8 | 28.0 | 3500.0 | — |
All figures are dense. Vendors commonly headline the sparse number, which is twice the dense one.
NVSwitch + NVLink