| Precision | Peak (PFLOPS) | Per GPU (TFLOPS) | TFLOPS/W |
|---|---|---|---|
| FP64 | 0.1 | 9.7 | 11.94 |
| FP32 | 0.2 | 19.5 | 24.00 |
| TF32 | 1.2 | 156.0 | 192.00 |
| FP16 | 2.5 | 312.0 | 384.00 |
| BF16 | 2.5 | 312.0 | 384.00 |
| INT8 | 5.0 | 624.0 | 768.00 |
All figures are dense. Vendors commonly headline the sparse number, which is twice the dense one.
NVSwitch + NVLink
DGX A100 with 640GB total GPU memory