NVIDIA H200 SXM5 141GB HBM3e GPU
Brand: NVIDIA | Category: GPUs
SKU: H200-SXM5-141GB | Part #: H200-SXM5-141GB | MPN: H200-SXM5-141GB
Contact for Pricing — Request a Quote
About the NVIDIA H200 SXM5 141GB HBM3e GPU
The NVIDIA H200 SXM5 141GB HBM3e GPU delivers 141 GB of HBM3e memory paired with an exceptional 4.8 TB/s memory bandwidth, setting a new standard for memory-intensive AI and scientific computing workloads. Built on the NVIDIA Hopper architecture (GH100), this enterprise-grade accelerator is engineered for data center deployment in the H200-SXM5-141GB configuration, combining massive on-chip memory capacity with the interconnect speeds required for real-time inference and large-scale model training.
Enterprise AI infrastructure teams and high-performance computing organizations choose the NVIDIA H200 for its unmatched tensor performance across multiple precision levels: 3,958 TFLOPS in FP8, 1,979 TFLOPS in both FP16 and BF16, and 989 TFLOPS in TF32—all with sparsity acceleration enabled. The integrated Transformer Engine supports mixed-precision compute in FP8 and FP16, allowing organizations to optimize throughput without sacrificing model accuracy. With 18 x NVLink 4.0 lanes delivering 900 GB/s bidirectional bandwidth per GPU, multi-GPU scaling becomes seamless. The addition of PCIe Gen 5 x16 connectivity, Up to 7 GPU instances via Multi-Instance GPU (MIG), full ECC memory support, and a 700 W thermal design power profile ensures reliability and efficiency across NVIDIA DGX H200 and NVIDIA HGX H200 platforms.
Whether you are architecting next-generation AI infrastructure or procuring GPUs for large language model training and deployment, the H200-SXM5-141GB addresses the most demanding performance and memory requirements. Omnixon Global supplies the full range of NVIDIA enterprise accelerators to organizations across the Middle East and beyond. Contact our team today with your specifications and volume requirements to request a formal quotation.
Technical Specifications
| Brand | NVIDIA |
| Category | GPUs |
| SKU | H200-SXM5-141GB |
| Part Number | H200-SXM5-141GB |
| Condition | New |
| Manufacturer Part Number | H200-SXM5-141GB |
| GPU Architecture | NVIDIA Hopper (GH100) |
| Form Factor | SXM5 |
| GPU Memory | 141 GB HBM3e |
| Memory Bandwidth | 4.8 TB/s |
| FP8 Tensor Core Performance | 3,958 TFLOPS (with Sparsity) |
| FP16 Tensor Core Performance | 1,979 TFLOPS (with Sparsity) |
| BF16 Tensor Core Performance | 1,979 TFLOPS (with Sparsity) |
| TF32 Tensor Core Performance | 989 TFLOPS (with Sparsity) |
| FP64 Tensor Core Performance | 67 TFLOPS |
| FP64 (IEEE) Performance | 34 TFLOPS |
| GPU Interconnect | NVLink 4.0, 900 GB/s bidirectional per GPU |
| NVLink Lanes | 18 x NVLink 4.0 |
| PCIe Interface | PCIe Gen 5 x16 |
| Thermal Design Power (TDP) | 700 W |
| Transformer Engine | Yes, with FP8 and FP16 mixed precision |
| Multi-Instance GPU (MIG) | Up to 7 GPU instances |
| ECC Memory | Yes |
| Platform Compatibility | NVIDIA DGX H200, NVIDIA HGX H200 |
Frequently Asked Questions about NVIDIA H200 SXM5 141GB HBM3e GPU
What does the NVIDIA H200 SXM5 141GB HBM3e GPU do?
The NVIDIA H200 SXM5 141GB HBM3e GPU accelerates AI/ML training, inference, scientific HPC, and virtualization (vGPU) workloads. Typical deployments include LLM training clusters, computer-vision pipelines, financial risk modeling, and rendering farms.
What are the headline specs of the NVIDIA H200 SXM5 141GB HBM3e GPU?
Key specifications for the NVIDIA H200 SXM5 141GB HBM3e GPU: new condition; raid support H200; gpu support NVIDIA H200; ai optimized Yes. Manufacturer part number H200-SXM5-141GB. For the full datasheet with electrical, environmental, and compliance details, contact our pre-sales engineering team.
Is the NVIDIA H200 SXM5 141GB HBM3e GPU compatible with my infrastructure?
The NVIDIA H200 SXM5 141GB HBM3e GPU requires a PCIe Gen4 or Gen5 x16 slot, server power adequate for the card's TDP, and CUDA/ROCm driver support in your hypervisor or bare-metal OS. Sales engineering will confirm chassis fit (1U/2U/4U), PCIe lane count, and PSU headroom before quoting.