Comparar
Fichas técnicas lado a lado para chips, sistemas e modelos. Cada dado mantém seu rótulo de origem — uma declaração do fabricante ao lado de uma estimativa é identificada exatamente como tal.
| NVIDIA H200 Hopper with 141 GB of HBM3E — the memory upgrade inference was waiting for. | NVIDIA H100 The GPU of the ChatGPT era — the unit of account for the 2023–25 AI build-out. | |
|---|---|---|
| Overview — Designer | NVIDIA | NVIDIA |
| Overview — Architecture | Hopper | Hopper (GH100) |
| Overview — Announced | November 2023; shipping from mid-2024 | March 2022 (GTC); shipping volume from early 2023 |
| Overview — Status | Shipping | Shipping — succeeded by Blackwell at the frontier |
| Manufacturing — Foundry / process | TSMC 4NVendor spec | TSMC 4N (custom 5 nm-class)Vendor spec |
| Memory — Memory | 141 GB HBM3EVendor spec | 80 GB HBM3 (SXM); 94 GB HBM3 (NVL); 80 GB HBM2e (PCIe)Vendor spec |
| Memory — Memory bandwidth | 4.8 TB/sVendor spec | 3.35 (SXM) / 3.9 (NVL) TB/sVendor spec |
| Compute — Compute | Same published tensor throughput as H100 (FP8 dense 1,979 TFLOPS)Vendor spec | — |
| Power — TDP | up to 700 WVendor spec | up to 700 (SXM) / 350 (PCIe) WVendor spec |
| Physical — Form factors | SXM5, NVL | SXM5, PCIe, NVL (paired PCIe) |
| Physical — Interconnect | NVLink 4 — 900 GB/sVendor spec | NVLink 4 — 900 GB/s; PCIe 5.0Vendor spec |
| Manufacturing — Transistors | — | 80 billionVendor spec |
| Manufacturing — Die size | — | 814 mm²Vendor spec |
| Manufacturing — Packaging | — | CoWoS-S with 5 or 6 HBM stacksVendor spec |
| Compute — TF32 tensor (dense) | — | 495 TFLOPSVendor spec |
| Compute — BF16/FP16 tensor (dense) | — | 990 TFLOPSVendor spec |
| Compute — FP8 tensor (dense) | — | 1,979 TFLOPSVendor spec |
| Compute — INT8 tensor (dense) | — | 1,979 TOPSVendor spec |
| Compute — FP64 tensor | — | 67 TFLOPSVendor spec |
| AI Performance — MLPerf | — | Set the reference results for training and inference benchmarks across 2023–24 submissionsMLPerf |
| Price — Launch price | — | Not publicly disclosed |
| Price — Market price | — | ~$25,000–40,000 per GPU at the 2023–24 peakReported |
| Price — Cloud rental | — | Roughly $2–7 per GPU-hour across major and specialist clouds during 2025 — falling steadilyReported |
"—" significa que esse registo não publica este campo; diferentes fornecedores divulgam informações distintas, e não preenchemos lacunas com suposições.