H100 vs H100 NVL: Specs
Architecture, compute capability and supported precisions for the H100 and the H100 NVL, from vendor datasheets.
No live price to compare
No provider on Aquanode is listing at least one of these two cards right now, so no hourly rate is shown. This page shows live prices as soon as both are listed.
Specs side by side
| Spec | H100 | H100 NVL |
|---|---|---|
| VRAM | 80GB HBM3 | 94GB HBM3 |
| Memory bandwidth | 3.35 TB/s | 3.9 TB/s |
| Architecture | Hopper | Hopper |
| Compute capability | 9.0 | 9.0 |
| Precisions in hardware | FP32, FP16, BF16, FP8, INT4 | FP32, FP16, BF16, FP8, INT4 |
| Peak FP16/BF16 tensor TFLOPS | 989 TFLOPS | 835.5 TFLOPS |
| Peak FP8 tensor TFLOPS | 1,979 TFLOPS | 1,670.5 TFLOPS |
| NVLink / multi-GPU interconnect | NVLink 4, 900 GB/s bidirectional | NVLink, 600 GB/s |
| Max board power (TDP) | Up to 700W (configurable) | 350-400W (configurable) |
