Product
NVIDIA H200 SXM 141GB
Part number NVIDIA H200 SXM
- Price
- USD $100
- Stock
- 1
Hopper accelerator with 141 GB of HBM3e memory at 4.8 TB/s, an SXM form factor, and a configurable TDP up to 700 W, designed for HGX H200 training and inference nodes.
- 141 GB HBM3e memory at 4.8 TB/s
- SXM form factor for NVIDIA HGX H200 systems
- NVLink at 900 GB/s plus PCIe Gen5
- Configurable TDP up to 700 W
- Up to seven MIG instances at 18 GB each
Specifications
| GPU Model | NVIDIA H200 SXM |
|---|---|
| Architecture | NVIDIA Hopper |
| Form Factor | SXM |
| GPU Count | 1 |
| GPU Memory | 141 GB HBM3e |
| Memory Bandwidth | 4.8 TB/s |
| Interconnect | NVLink 900 GB/s; PCIe Gen5 128 GB/s |
| Max TDP | Up to 700 W (configurable) |
| Cooling | Host-cooled SXM module |
| FP64 | 34 TFLOPS |
| FP64 Tensor Core | 67 TFLOPS |
| FP32 | 67 TFLOPS |
| TF32 Tensor Core | 989 TFLOPS (with sparsity) |
| FP16 / BFLOAT16 Tensor Core | 1,979 TFLOPS (with sparsity) |
| FP8 Tensor Core | 3,958 TFLOPS (with sparsity) |
| INT8 Tensor Core | 3,958 TOPS (with sparsity) |
| Multi-Instance GPU | Up to 7 MIGs at 18 GB each |
| Confidential Computing | Supported |
| Media Engines | 7 NVDEC, 7 JPEG |
| Typical Host | NVIDIA HGX H200 (4- or 8-GPU) partner and NVIDIA-Certified Systems |
Description
The NVIDIA H200 SXM is the Hopper-generation successor to H100 in high-density SXM servers, distinguished mainly by a much larger, faster memory subsystem. NVIDIA describes it as the first data-center GPU to use HBM3e, with 141 GB of memory at 4.8 TB/s — nearly double the capacity of H100 and about 1.4 times the memory bandwidth.
Compute rates published for the SXM variant match the H100 SXM class, including 34 TFLOPS of FP64 and, with sparsity, 1,979 TFLOPS of FP16/BF16 Tensor performance and 3,958 TFLOPS of FP8 Tensor performance. Fourth-generation NVLink remains at 900 GB/s, with PCIe Gen5 at 128 GB/s to the host. In an eight-GPU HGX H200 board, NVSwitch provides a fully connected fabric across the modules. NVIDIA also lists confidential computing support and up to seven MIG instances at 18 GB each.
Thermal design power is configurable up to 700 W. As with other SXM modules, cooling is provided by the host, which may be air or liquid depending on the chassis. The larger memory window is the practical reason to choose H200 over H100: longer context, larger models, and fewer shards for inference, without leaving the Hopper software stack.
This module is the GPU used in NVIDIA HGX H200 partner systems such as the GIGABYTE G593-SD1. The related H200 NVL card offers the same 141 GB HBM3e capacity in a PCIe, air-cooled form factor. Request a quote for the SXM module or a complete eight-GPU host.
Related products
GIGABYTE G593-SD1-AAX3
P/N G593-SD1-AAX3
Price USD $100 Stock 1
5U dual-socket Intel Xeon server built around an NVIDIA HGX H200 eight-GPU SXM module, with 900 GB/s NVLink/NVSwitch bandwidth and 4+2 3000 W Titanium power supplies.
- Model
- G593-SD1-AAX3
- GPU Model
- NVIDIA HGX H200
- GPU Count
- 8 SXM
- Form Factor
- 5U rackmount
NVIDIA H200 NVL 141GB
P/N NVIDIA H200 NVL
Price USD $100 Stock 1
Hopper PCIe accelerator with 141 GB of HBM3e memory at 4.8 TB/s, a dual-slot air-cooled form factor, and a configurable TDP up to 600 W, for MGX and NVIDIA-Certified servers.
- GPU Model
- NVIDIA H200 NVL
- Form Factor
- PCIe dual-slot air-cooled
- GPU Count
- 1
- GPU Memory
- 141 GB HBM3e
Ready to specify a configuration?
Share the use case, the target build, quantity, destination country, and timeline. We will take it from the inquiry.