| Precision | Peak (PFLOPS) | Per GPU (TFLOPS) | TFLOPS/W |
|---|---|---|---|
| FP64 | 0.3 | 34.0 | 34.00 |
| FP32 | 0.5 | 67.0 | 67.00 |
| TF32 | 4.0 | 494.5 | 989.00 |
| FP16 | 7.9 | 989.5 | 1979.00 |
| BF16 | 7.9 | 989.5 | 1979.00 |
| FP8 | 15.8 | 1979.0 | 3958.00 |
| INT8 | 15.8 | 1979.0 | 3958.00 |
All figures are dense. Vendors commonly headline the sparse number, which is twice the dense one.
NVSwitch + NVLink
HGX H100 8-GPU configuration for hyperscalers