| Product SKU | H100-SXM5-80GB |
|---|---|
| FP64 | 34 teraFLOPS |
| FP64 Tensor Core | 67 teraFLOPS |
| FP32 | 67 teraFLOPS |
| TF32 Tensor Core * | 989 teraFLOPS |
| BFLOAT16 Tensor Core * | 1,979 teraFLOPS |
| FP16 Tensor Core * | 1,979 teraFLOPS |
| FP8 Tensor Core * | 3,958 teraFLOPS |
| INT8 Tensor Core * | 3,958 TOPS |
| GPU Memory | 80GB |
| GPU Memory Bandwidth | 3.35TB/s |
| Decoders | 7 NVDEC 7 JPEG |
| Max Thermal Design Power (TDP) | Up to 700W (configurable) |
| Multi-Instance GPUs | Up to 7 MIGS @ 10GB each |
| Form Factor | SXM |
| Interconnect | NVIDIA NVLink™: 900GB/s PCIe Gen5: 128GB/s |
| Server Options | NVIDIA HGX H100 Partner and NVIDIA- Certified Systems ™ with 4 or 8 GPUs NVIDIA DGX H100 with 8 GPUs |
| NVIDIA AI Enterprise | Add-on |
Data Center GPUs, GPU
NVIDIA H100 SXM5 80GB
The NVIDIA H100 SXM5 80GB is a data-center GPU, with 80GB of memory and 3.35TB/s of memory bandwidth. It combines 3,958 teraFLOPS FP8 and 1,979 teraFLOPS FP16 Tensor Core performance with a board power of up to 700 W (configurable) in a SXM form factor.
Overview
The NVIDIA H100 SXM5 80GB is a data-center GPU built on the NVIDIA Hopper architecture for large-scale AI training, fine-tuning and HPC. It pairs 80 GB of HBM3 with 3.35 TB/s of memory bandwidth and Tensor Core peaks of 3,958 TFLOPS FP8 and 1,979 TFLOPS FP16/BF16. The SXM5 form factor and 900 GB/s NVLink interconnect are intended for dense HGX H100 4- and 8-GPU platforms, where the GPUs operate as one node rather than as independent add-in cards.
At a glance
- Architecture: NVIDIA Hopper; 80 GB HBM3 at 3.35 TB/s
- Tensor Core peaks (with sparsity): 3,958 TFLOPS FP8 · 1,979 TFLOPS FP16/BF16 · 989 TFLOPS TF32
- FP64/FP32: 34 TFLOPS FP64 · 67 TFLOPS FP64 Tensor Core · 67 TFLOPS FP32
- Interconnect: NVLink 900 GB/s; PCIe Gen5 128 GB/s
- Form factor: SXM5; TDP up to 700 W (configurable)
- Multi-Instance GPU: up to 7 MIG instances of 10 GB
- Confidential Computing; 7 NVDEC and 7 JPEG decoders
- Platforms: NVIDIA HGX H100 (4 or 8 GPUs) and DGX H100 (8 GPUs)
When to choose the H100 SXM5
Choose the SXM5 version when the workload needs maximum GPU-to-GPU bandwidth and the server is built for it: LLM training and fine-tuning, large-batch inference and tightly coupled HPC simulation all benefit from NVLink. If the deployment is limited to standard PCIe slots or a lower board power, a PCIe-class card of the same generation is the better fit.
Platform & compatibility
H100 SXM5 modules mount on NVIDIA HGX H100 partner systems with four or eight GPUs and on DGX H100 systems with eight GPUs; they are not a standalone add-in card. The platform is paired with NVIDIA AI Enterprise software for a supported AI stack, available as an add-on.
FAQ
Is the H100 SXM5 a plug-in PCIe card? No — it is an SXM5 module that mounts on an HGX or DGX baseboard. For standard PCIe servers, use an H100 PCIe or NVL card.
How much memory does it have? 80 GB of HBM3 with 3.35 TB/s of bandwidth.
Can it be shared between workloads? Yes. Multi-Instance GPU (MIG) can partition it into up to seven instances of 10 GB.
Related
- All NVIDIA data-center GPUs
- Data-center GPU buying guide: A100 to B300
- Browse all GPUs and accelerators
Configuration, availability and lead time depend on the required quantity, server pairing and delivery destination. Send your requirement to receive a quotation.
