Nvidia H100 SXM
The Nvidia H100 SXM is Nvidia’s Hopper-generation data-center GPU with 80 GB of HBM3.
- Vendor
- Nvidia
- Architecture
- Hopper
- Memory
- 80 GB HBM3
- Memory bandwidth (TB/s)
- 3.35 TB/s
- FP8 (TFLOPS)
- 3,958 TFLOPS
- Power (W)
- 700 W
- Form factor
- SXM module (HGX H100 / DGX H100)
- Status
- Shipping
The SXM version listed here goes into 8-GPU HGX H100 baseboards and DGX H100 systems; Nvidia also sells a PCIe H100 NVL card with 94 GB and 3.9 TB/s. Nvidia’s product page gives 80 GB, 3.35 TB/s and 3,958 TFLOPS FP8 with sparsity at up to 700 W; Nvidia’s HGX reference architecture lists the memory as HBM3.
- Memory (GB)
- 80 GB
- Precision note (sparse or dense)
- 3,958 TFLOPS FP8 Tensor Core is stated with sparsity on Nvidia's H100 page; Nvidia's Blackwell Ultra chip table gives Hopper FP8 as 2 PFLOPS dense / 4 sparse. H100 NVL: 3,341 TFLOPS with sparsity.
- Vendor source
- nvidia.com
- Checked
Questions
How much memory does the Nvidia H100 have?
The H100 SXM has 80 GB of HBM3 at 3.35 TB/s; the PCIe H100 NVL card has 94 GB at 3.9 TB/s.
What is the H100’s FP8 performance?
Nvidia’s H100 page gives 3,958 TFLOPS FP8 with sparsity; Nvidia’s Blackwell Ultra chip table gives Hopper FP8 as 2 PFLOPS dense and 4 sparse.
How much power does the H100 use?
Nvidia lists the H100 SXM at up to 700 W.
What is the difference between H100 SXM and H100 NVL?
The SXM version goes into HGX H100 and DGX H100 systems with 80 GB; the H100 NVL is a PCIe card with 94 GB and 3,341 TFLOPS FP8 with sparsity.
Source
Every figure is taken from the vendor’s own datasheet, product page, technical documentation or announcement, linked on each chip with the date it was last checked. How we pick and check sources: editorial standards.