| Precision | Peak (TFLOPS) | Per GPU (TFLOPS) | TFLOPS/W |
|---|---|---|---|
| FP64 | 272 | 34 | 64.00 |
| FP32 | 536 | 67 | 126.12 |
| TF32 | 3,956 | 495 | 930.82 |
| FP16 | 7,916 | 990 | 1862.59 |
| BF16 | 7,916 | 990 | 1862.59 |
| FP8 | 15,832 | 1,979 | 3725.18 |
| INT8 | 15,832 | 1,979 | 3725.18 |
All figures are dense. Vendors commonly headline the sparse number, which is twice the dense one.
NVSwitch + NVLink
Latest DGX system with H200 GPUs and 141GB HBM3e memory per GPU