| Product SKU | H200-SXM5-141GB |
|---|---|
| FP64 | 34 TFLOPS |
| FP64 Tensor Core | 67 TFLOPS |
| FP32 | 67 TFLOPS |
| TF32 Tensor Core² | 989 TFLOPS |
| BFLOAT16 Tensor Core² | 1,979 TFLOPS |
| FP16 Tensor Core² | 1,979 TFLOPS |
| FP8 Tensor Core² | 3,958 TFLOPS |
| INT8 Tensor Core² | 3,958 TFLOPS |
| GPU Memory | 141GB |
| GPU Memory Bandwidth | 4.8TB/s |
| Decoders | 7 NVDEC 7 JPEG |
| Confidential Computing | Supported |
| Max Thermal Design Power (TDP) | Up to 700W (configurable) |
| Multi-Instance GPUs | Up to 7 MIGs @18GB each |
| Form Factor | SXM |
| Interconnect | NVIDIA NVLink™: 900GB/s PCIe Gen5: 128GB/s |
| Server Options | NVIDIA HGX™ H200 partner and NVIDIA-Certified Systems™ with 4 or 8 GPUs |
| NVIDIA AI Enterprise | Add-on |
Data Center GPUs, GPU
NVIDIA H200 SXM5 141GB
The NVIDIA H200 SXM5 141GB is a data-center GPU, with 141GB of memory and 4.8TB/s of memory bandwidth. It combines 3,958 TFLOPS FP8 and 1,979 TFLOPS FP16 Tensor Core performance with a board power of up to 700 W (configurable) in a SXM form factor.
Overview
The NVIDIA H200 SXM5 141GB is a data-center GPU for large-scale AI training, fine-tuning and inference. Built on the Hopper architecture, it raises memory to 141 GB of HBM3e at 4.8 TB/s — the key upgrade over the H100 SXM5 — while keeping the same Tensor Core class of 3,958 TFLOPS FP8 and 1,979 TFLOPS FP16/BF16. Its SXM5 form factor and 900 GB/s NVLink target dense HGX H200 4- and 8-GPU platforms.
At a glance
- Architecture: NVIDIA Hopper; 141 GB HBM3e at 4.8 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 18 GB
- Confidential Computing; 7 NVDEC and 7 JPEG decoders
- Platforms: NVIDIA HGX H200 (4 or 8 GPUs); AI Enterprise add-on
When to choose the H200 SXM5
Choose the H200 SXM5 when the workload is limited by memory capacity or bandwidth rather than raw compute: very large language models, long-context inference and large-scale fine-tuning benefit most from the 141 GB of HBM3e at 4.8 TB/s. Where 80 GB is sufficient and budget is tighter, the H100 SXM5 remains the workhorse.
Platform & compatibility
H200 SXM5 modules mount on NVIDIA HGX H200 partner systems with four or eight GPUs; they are not a standalone add-in card. The platform pairs with NVIDIA AI Enterprise software (add-on), and the module supports Confidential Computing and Multi-Instance GPU.
FAQ
How is it different from the H100 SXM5? Same compute class, but 141 GB HBM3e at 4.8 TB/s versus 80 GB HBM3 at 3.35 TB/s.
Is it a PCIe card? No — it is an SXM5 module for HGX H200 baseboards.
Can it be partitioned? Yes — up to seven MIG instances of 18 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.
