NVIDIA H100 PCIe 80GB
TrainingNvidia · Hopper · 2022
PCIe form factor; up to 2 GPUs bridged via NVLink (600 GB/s). Memory bandwidth >2 TB/s (lower than SXM5's >3.35 TB/s). fp16Tflops/fp8Tflops are NVIDIA's published Tensor Core peak figures "with sparsity" (dense throughput is approximately half); see source.
- VRAM
- 80 GB
- Memory
- HBM2e
- FP16
- 1513 TF
- FP8
- 3026 TF
From
Verified 8 Jul 2026€1.89/GPU·h
Offerings of this GPU
4 rows
| Provider | Config | Regions | €/GPU·h | Availability | Sov. | Verified | |
|---|---|---|---|---|---|---|---|
| Seeweb Compute | 1× GPU · VM Cloud Server GPU - H100 | 🇮🇹 IT | €1.89/GPU·h | Available | 63 | 2026-07-08 | |
| Scaleway Compute | 2× GPU · VM H100-2-80G | 🇫🇷 FR🇵🇱 PL | €2.73/GPU·h egress free | Available | 69 | 2026-07-08 | |
| OVHcloud Compute | 1× GPU · VM Public Cloud GPU H100 (entry tier) | 🇫🇷 FR🇩🇪 DE | €2.80/GPU·h egress free | Available | 80 | 2026-07-08 | |
| OVHcloud Inference API | 1× GPU · Serverless Managed AI (H100 PCIe endpoint) | 🇫🇷 FR | Quote only egress free | Available | 80 | 2026-07-08 |
Spec sources
- Official Docswww.nvidia.com/content/dam/en-zz/Solutions/gtcs22/data-center/h100/PB-11133-001_v01.pdf8 Jul 2026
- Official Docswww.nvidia.com/en-us/data-center/h1008 Jul 2026