| Main Specifications | |
| Product Series | Nvidia H100 |
| Core Type | NVIDIA TENSOR |
| Host Interface | PCI Express 5.0 x16 |
| GPU Architecture | Hopper |
| Detailed Specifications | |
| PCIe x16 Interconnect Bandwidth | PCIe Gen5: 128GB/s |
| Max Memory Size | 80 GB |
| Max Memory Bandwidth | 2.0TB/s |
| Peak FP64 | 24 teraFLOPS |
| Peak FP64 Tensor Core | 48 teraFLOPS |
| INT8 Tensor Core | 3,200 TOPS |
| TF32 Tensor Core | 800 teraFLOPS |
| Peak Single Precision (FP32) Performance | 48 teraFLOPS |
| Peak BFLOAT16 Tensor Core | 1,600 teraFLOPS |
| Peak FP16 Tensor Core | 1,600 teraFLOPS |
| Peak FP8 Tensor Core | 3,200 teraFLOPS |
| Total NVLink Bandwidth | 600GB/s |
| Multi-Instance GPUs | Up to 7 MIGS @ 10GB each |
| Cooling | Passive |
| Dual Slot | Yes |
| Form Factor | PCIe |
| Supplementary Power Connectors | 1x 16-pin PCIe CEM5 |
| Max Graphics Card Power (W) | 350W |
