NVIDIAAI GPU · PCIe
H100 PCIe
NVIDIA H100 PCIe is a Hopper data-center accelerator with 80 GB HBM2e and 2.04 TB/s of memory bandwidth.
Specifications
Dense throughput, no sparsity| Architecture | Hopper |
|---|---|
| Form factor | PCIe |
| Process | TSMC 4N |
| FP16 / BF16 tensor (dense) | 756 TFLOPS |
| FP8 tensor (dense) | — |
| FP4 tensor (dense) | — |
| Memory | 80 GB HBM2e |
| Memory bandwidth | 2.04 TB/s |
| Power | 350 W |
| Launch | Sep 2022 |
AI index
Tensor × bandwidth100 × √((756 ÷ 989.4) × (2.04 ÷ 3.35)) = 68.2
The AI index is the geometric mean of dense 16-bit tensor throughput and memory bandwidth, each relative to the H100 SXM. It is a spec-sheet comparison, not a measured benchmark.