Key Architecture Highlights
- Specifications Version: The H200 NVL pre-specification currently listed on the official website, which manufacturers note may be adjusted
- GPU Architecture: NVIDIA Hopper
- GPU memory specifications: GPU memory: 141GB; GPU memory Type: HBM3e
Applications
- Big language model reasoning
- Artificial intelligence
- High-performance computing
Engineering Parameters
| Specifications Version | The H200 NVL pre-specification currently listed on the official website, which manufacturers note may be adjusted |
|---|---|
| GPU Architecture | NVIDIA Hopper |
| GPU memory specifications | GPU memory: 141GB GPU memory Type: HBM3e |
| GPU memory bandwidth | 4.8TB/s |
| General-purpose compute performance | FP32: 60 TFLOPS FP64: 30 TFLOPS FP64 Tensor Core: 60 TFLOPS |
| Tensor Compute performance | TF32 Tensor Core: 835 TFLOPS (Sparse) FP16 Tensor Core: 1, 671 TFLOPS (Sparse) BF16 Tensor Core: 1, 671 TFLOPS (Sparse) FP8 Tensor Core: 3, 341 TFLOPS (Sparse) |
| NVLink / Interconnection | Connectivity: 2- or 4-way NVIDIA NVLink bridge: 900GB/s per GPU PCIe Gen5: 128GB/s |
| MIG Division | The maximum number of instances is 7 instances, with a maximum of 16.5GB per instance. |
| Video decoding | 7 NVDEC/7 JPEG decoders |
| Board power | Maximum 600W configurable |
| Card form factor / cooling | Card form factor: PCIe Dual-slot air-cooled |
Warranty
manufacturer's public product information does not specify the warranty period for this model in its supply channel; applicable region, channel and actual warranty service confirmed by the supply contract.
Manufacturer information
Manufacturer informationVerified Documentation & Datasheets
Official products page
Download DocumentNVIDIA H200 Tensor Core GPU | Datasheet
Download Document