NVIDIA H200 PCIe 141GB – Hopper Tensor Core Data Center GPU
Home / Memory / NVIDIA H200 PCIe 141GB – Hopper Tensor Core Data Center GPU

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

The NVIDIA H200 PCIe 141GB is a high-performance Hopper-generation data-center GPU accelerator engineered for artificial intelligence, generative AI, large language models, high-performance computing (HPC), scientific computing, and accelerated data analytics. It features 141GB of HBM3e memory, up to 4.8TB/s memory bandwidth, 16,896 CUDA cores, fourth-generation Tensor Cores, and a PCIe Gen5 x16 host interface.

Category:

Description

The NVIDIA H200 PCIe 141GB extends NVIDIA’s Hopper architecture into a PCIe-based data-center accelerator, providing a large high-bandwidth memory pool for workloads that require substantial GPU memory capacity and throughput.

Its 141GB HBM3e memory provides up to 4.8TB/s of memory bandwidth, allowing large AI models, datasets, and intermediate computations to remain in high-speed GPU memory. This is particularly valuable for large language models, generative AI, inference, scientific simulations, and other memory-intensive workloads.

The accelerator incorporates 16,896 CUDA cores and 528 fourth-generation Tensor Cores, providing extensive parallel compute resources for AI and HPC applications. Tensor Core acceleration supports modern AI numerical formats including FP64, FP32, TF32, FP16, BF16, FP8, and INT8.

Unlike the H200 SXM module, the H200 PCIe is installed as a PCI Express expansion accelerator. It uses a PCIe Gen5 x16 interface and is designed as a dual-slot, air-cooled data-center card, allowing deployment in compatible PCIe server platforms without requiring an SXM-specific GPU baseboard.

The H200 PCIe also supports NVIDIA Multi-Instance GPU (MIG) technology, allowing the GPU’s resources to be partitioned into isolated GPU instances for workloads requiring GPU sharing and resource isolation.

For multi-GPU deployments, the H200 PCIe supports high-speed GPU interconnect technologies appropriate to the platform configuration. System-level compatibility, power, cooling, and supported interconnect configuration should be checked with the server manufacturer before deployment.

3. Complete Technical Specification

Specification Details
Manufacturer NVIDIA
Product H200 Tensor Core GPU
Variant H200 PCIe 141GB
Architecture NVIDIA Hopper
GPU Architecture GH100
Form Factor PCIe dual-slot
Cooling Air-cooled
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
Host Interface PCI Express Gen5 x16
PCIe Generation PCIe 5.0
PCIe Width x16
MIG Supported
Maximum MIG Instances Up to 7
GPU Interconnect Platform-dependent high-speed GPU interconnect
Power Envelope Approximately 600W-class
Cooling Design Data-center air cooling
Primary Deployment Enterprise data center / HPC
Primary Workloads AI, ML, Generative AI, LLM, HPC, analytics

* NVIDIA’s published Tensor Core peak figures with sparsity are theoretical maximums. Actual application performance varies according to workload, precision, software, and system configuration.

4. Applications

The H200 PCIe 141GB is designed for demanding accelerated-computing applications, including:

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

5. Compatibility

The H200 PCIe 141GB is designed for PCIe-based data-center servers equipped to support high-power, dual-slot NVIDIA GPU accelerators.

A compatible host system should provide:

  • PCIe Gen5 x16 slot or qualified compatible PCIe configuration
  • Adequate dual-slot GPU clearance
  • Sufficient GPU power delivery
  • Appropriate auxiliary power connectivity
  • Data-center-class airflow and cooling
  • Compatible BIOS/UEFI and server firmware
  • NVIDIA H200-supported drivers
  • Sufficient chassis and PSU capacity
  • Server manufacturer’s approval for the H200 PCIe configuration

Important Compatibility Note

The H200 PCIe is not interchangeable with the H200 SXM module.

The PCIe model uses a conventional PCIe accelerator-card interface, while the SXM model requires an SXM-compatible GPU baseboard and system architecture.

For reliable deployment, verify the exact server manufacturer’s GPU support matrix before installation.

6. Key Benefits

  • 141GB HBM3e GPU memory
  • Up to 4.8TB/s memory bandwidth
  • 16,896 CUDA cores
  • 528 fourth-generation Tensor Cores
  • NVIDIA Hopper architecture
  • PCIe Gen5 x16 host interface
  • High memory capacity for large AI models
  • Excellent performance for LLM workloads
  • Optimized for generative AI and inference
  • Support for Multi-Instance GPU (MIG)
  • Up to seven MIG instances
  • Dual-slot air-cooled PCIe design
  • Suitable for enterprise data-center servers
  • Ideal for HPC and scientific workloads
  • Designed for memory-intensive AI applications
  • Easier PCIe server integration than SXM-based accelerators

7. SEO Title

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

Related products