Official H100 family artwork showing an SXM module. PCIe NVL variants have a different physical form.
Official H100 family artwork showing an SXM module. PCIe NVL variants have a different physical form.

NVIDIA / Accelerator Family

NVIDIA H100 Tensor Core GPU

Hopper accelerator family for AI training, inference and scientific computing. The SXM and NVL variants differ in memory, power and platform requirements.

Include your preferred configuration and quantity in your enquiry.

Overview

Hopper accelerator family for AI training, inference and scientific computing. The SXM and NVL variants differ in memory, power and platform requirements.

Product highlights

  • SXM variant: 80GB; up to 700W configurable TDP
  • NVL variant: 94GB per GPU; 350–400W configurable TDP
  • SXM memory bandwidth: 3.35TB/s

Platform specifications

Specifications describe model options or platform capabilities. Confirm the exact part number, configuration and regional availability when requesting a quotation.

NVIDIA H100 Tensor Core GPU specifications
SXM variant80GB; up to 700W configurable TDP
NVL variant94GB per GPU; 350–400W configurable TDP
SXM memory bandwidth3.35TB/s
NVL memory bandwidth3.9TB/s per GPU
ArchitectureNVIDIA Hopper

Model variants

H100 SXM
Interface: SXM module; compatible baseboard required
H100 NVL
Interface: PCIe; compatible server required

Specify the required variant when requesting a quote.

Official product resources

Explore the same category.

View Graphics Cards