GPU Wiki Series
NVIDIA H-Series
Hopper data center GPU family. Architecture coverage: Hopper.
Quick Facts
- Architecture
- Hopper
- Launch range
- 2022-2024
- Memory types
- HBM2e, HBM3, HBM3e
GPU entries by product stack
Desktop and laptop GPUs are listed in product-stack order.
| GPU | Architecture | Launch Date | Shader Processors | Memory | Bus Width | TDP/TBP | Segment |
|---|---|---|---|---|---|---|---|
| H100 CNX | Hopper | 2022-03-22 | 14,592 | 80 GB HBM2e | 5,120-bit | 350 W | Data Center |
| H100 NVL 94 GB | Hopper | 2023-03-21 | 16,896 | 94 GB HBM3 | 6,016-bit | 400 W | Data Center |
| H100 PCIe 80 GB | Hopper | 2022-03-22 | 14,592 | 80 GB HBM2e | 5,120-bit | 350 W | Data Center |
| H100 PCIe 96 GB | Hopper | 2023-03-21 | 16,896 | 96 GB HBM3 | 5,120-bit | 700 W | Data Center |
| H100 SXM5 64 GB | Hopper | 2023-03-21 | 16,896 | 64 GB HBM3 | 3,072-bit | 700 W | Data Center |
| H100 SXM5 80 GB | Hopper | 2022-03-22 | 16,896 | 80 GB HBM3 | 5,120-bit | 700 W | Data Center |
| H100 SXM5 94 GB | Hopper | 2023-03-21 | 16,896 | 94 GB HBM3 | 5,120-bit | 700 W | Data Center |
| H100 SXM5 96 GB | Hopper | 2023-03-21 | 16,896 | 96 GB HBM3 | 5,120-bit | 700 W | Data Center |
| H200 NVL | Hopper | 2024-11-18 | 16,896 | 141 GB HBM3e | 6,144-bit | 600 W | Data Center |
| H200 SXM 141 GB | Hopper | 2023-11-13 | 16,896 | 141 GB HBM3e | 6,144-bit | 700 W | Data Center |
| H800 PCIe 80 GB | Hopper | 2023-03-21 | 14,592 | 80 GB HBM2e | 5,120-bit | 350 W | Data Center |
| H800 SXM5 | Hopper | 2023-03-21 | 16,896 | 80 GB HBM3 | 5,120-bit | 700 W | Data Center |
Series Notes
NVIDIA H-Series became a major AI training and inference generation through H100 and H200 class products. Platform, interconnect and memory capacity define the real deployment.