NVIDIA H200 SXM 141GB – Hopper Tensor Core Data Center GPU
NVIDIA H200 SXM 141GB – Hopper Tensor Core Data Center GPU - Image 2
NVIDIA H200 SXM 141GB – Hopper Tensor Core Data Center GPU - Image 3
Home / Memory / NVIDIA H200 SXM 141GB – Hopper Tensor Core Data Center GPU

NVIDIA H200 SXM 141GB – Hopper Tensor Core Data Center GPU

The NVIDIA H200 SXM 141GB is a high-performance Hopper-generation data-center GPU accelerator designed for large-scale artificial intelligence, generative AI, high-performance computing (HPC), and scientific workloads. It features 141GB of HBM3e memory, up to 4.8TB/s memory bandwidth, 16,896 CUDA cores, fourth-generation Tensor Cores, and high-speed NVIDIA NVLink connectivity for scalable multi-GPU computing.

Category:

Description

The NVIDIA H200 SXM 141GB is an advanced data-center GPU based on the NVIDIA Hopper architecture. It was designed to address increasingly memory-intensive AI and HPC workloads, particularly applications involving large language models, generative AI, scientific simulations, and large-scale inference.

The major enhancement over the H100 generation is the substantially larger and faster GPU memory subsystem. The H200 provides 141GB of HBM3e memory with up to 4.8TB/s of memory bandwidth, allowing large models and datasets to remain resident in high-bandwidth GPU memory while reducing the need for frequent data transfers.

The GPU incorporates 16,896 CUDA cores and 528 fourth-generation Tensor Cores, providing massive parallel processing capability for AI and accelerated computing. NVIDIA’s Tensor Core architecture supports modern precision formats including FP64, FP32, TF32, FP16, BF16, FP8, and INT8.

For multi-GPU environments, the H200 SXM supports NVIDIA NVLink, enabling high-speed GPU-to-GPU communication. This is particularly important in systems containing multiple accelerators where distributed AI training and inference workloads require rapid exchange of model parameters and intermediate data.

The H200 SXM is designed as a server-integrated SXM module, rather than a conventional PCIe graphics card. It is intended for purpose-built platforms such as NVIDIA HGX H200 systems and qualified OEM data-center servers with the appropriate power delivery, cooling, system firmware, and NVLink infrastructure.

The accelerator also supports Multi-Instance GPU (MIG) technology, allowing compatible workloads to partition the GPU into multiple isolated GPU instances for improved resource utilization.

3. Complete Technical Specification

Specification Details
Manufacturer NVIDIA
Product H200 Tensor Core GPU
Variant H200 SXM
Architecture NVIDIA Hopper
GPU Architecture GH100
Form Factor SXM
GPU Memory 141GB
Memory Type HBM3e
Memory Bandwidth Up to 4.8TB/s
CUDA Cores 16,896
Tensor Cores 528 fourth-generation Tensor Cores
FP64 Performance Up to 34 TFLOPS
FP64 Tensor Core Up to 67 TFLOPS
FP32 Performance Up to 67 TFLOPS
TF32 Tensor Core Up to 989 TFLOPS*
BF16 Tensor Core Up to 1,979 TFLOPS*
FP16 Tensor Core Up to 1,979 TFLOPS*
FP8 Tensor Core Up to 3,958 TFLOPS*
INT8 Tensor Core Up to 3,958 TOPS*
L2 Cache 50MB
GPU Interconnect NVIDIA NVLink
NVLink Bandwidth Up to 900GB/s
MIG Supported
Maximum MIG Instances Up to 7
TDP Up to approximately 700W
Host Interface SXM platform interface
Memory Technology HBM3e
Primary Deployment Data center / accelerated computing
Typical Configuration Multi-GPU HGX / enterprise server platforms
Primary Workloads AI, ML, Generative AI, HPC, scientific computing

* Tensor Core figures are NVIDIA’s peak theoretical figures with sparsity where applicable; actual application performance varies by workload and software.

4. Applications

The H200 SXM 141GB is designed for demanding workloads including:

  • Large Language Model (LLM) training
  • Generative AI
  • Large-scale AI inference
  • Deep learning
  • Transformer-based models
  • Natural language processing
  • Computer vision
  • Recommendation systems
  • High-performance computing (HPC)
  • Scientific simulations
  • Computational fluid dynamics
  • Molecular and pharmaceutical research
  • Financial modeling
  • Seismic and geophysical processing
  • GPU-accelerated analytics
  • Digital twins
  • AI supercomputing
  • Multi-GPU distributed computing
  • Enterprise AI infrastructure

5. Compatibility

The H200 SXM 141GB is intended for server platforms specifically engineered for NVIDIA SXM accelerators.

Compatible deployment environments include:

  • NVIDIA HGX H200
  • Qualified OEM H200 server platforms
  • Enterprise AI servers designed for H200 SXM modules
  • Multi-GPU accelerated computing systems

A compatible system must provide the appropriate:

  • SXM GPU interface
  • High-capacity GPU power delivery
  • Data-center-class cooling
  • System firmware/BIOS support
  • NVIDIA H200-compatible software and drivers
  • NVLink infrastructure for multi-GPU operation
  • Chassis clearance and thermal design
  • Power supply capacity appropriate for the complete server configuration

Important Compatibility Note

The H200 SXM is not a standard PCIe graphics card. It cannot simply be installed into a conventional PCIe x16 expansion slot.

For store catalog purposes, it should be described as an SXM data-center GPU accelerator module intended for compatible HGX/OEM server platforms.

6. Key Benefits

  • 141GB HBM3e high-bandwidth GPU memory
  • Up to 4.8TB/s memory bandwidth
  • 16,896 CUDA cores
  • 528 fourth-generation Tensor Cores
  • Hopper architecture optimized for AI and HPC
  • Exceptional memory capacity for large AI models
  • High-speed NVLink connectivity
  • Up to 900GB/s NVLink bandwidth
  • Support for Multi-Instance GPU (MIG)
  • Up to seven MIG instances
  • Excellent platform for LLM training and inference
  • Designed for generative AI workloads
  • Suitable for scientific and engineering computing
  • Optimized for multi-GPU data-center configurations
  • Up to approximately 700W GPU power envelope
  • Purpose-built for enterprise AI infrastructure

7. SEO Title

NVIDIA H200 SXM 141GB HBM3e Tensor Core GPU – 4.8TB/s AI & HPC Accelerator

Related products