GPU/GPU Stack/NVIDIA H200
Latest Generation

NVIDIA H200

The world's most powerful GPU for large language model training. 141GB of HBM3e memory and 4.8 TB/s bandwidth.

Key Specifications

ArchitectureHop
VRAM141 GB HBM3e
Memory Bandwidth4.8 TB/s
CUDA Cores16,896
Tensor Cores528
FP16 Performance1,979 TFLOPS

Technical Specifications

ArchitectureHop
VRAM141 GB HBM3e
Memory Bandwidth4.8 TB/s
CUDA Cores16,896
Tensor Cores528
FP16 Performance1,979 TFLOPS
FP8 Performance3,958 TFLOPS
TDP700W
InterconnectNVLink 4.0 (900 GB/s)
PCIeGen5 x16
Form FactorSXM5

Pricing Plans

Flexible pricing options to match your workload requirements.

On-Demand

Pay as you go with no commitment

800/hour
  • 1x NVIDIA H200 GPU
  • 64 vCPUs
  • 512 GB RAM
  • 2 TB NVMe SSD
  • No minimum commitment
  • Start/stop anytime
Most Popular

Reserved 1 Month

Save 15% with monthly commitment

340,000/month
  • 1x NVIDIA H200 GPU
  • 64 vCPUs
  • 512 GB RAM
  • 2 TB NVMe SSD
  • 15% discount
  • Priority support

Reserved 1 Year

Maximum savings with annual commitment

240,000/month
  • 1x NVIDIA H200 GPU
  • 64 vCPUs
  • 512 GB RAM
  • 2 TB NVMe SSD
  • 40% discount
  • Dedicated support

Why Choose NVIDIA H200

Extreme Bandwidth

4.8 TB/s memory bandwidth for faster data movement and reduced bottlenecks.

Unprecedented Memory

141GB HBM3e - 76% more memory than H100 for larger models and batch sizes.

NVLink 4.0

900 GB/s GPU-to-GPU bandwidth for efficient multi-GPU scaling.

Transformer Engine

Automatic mixed precision with FP8 support for 2x throughput on transformers.

Use Cases

Multi-Modal AI

Build vision-language models requiring massive memory for image and text processing.

Large Language Model Training

Train models with hundreds of billions of parameters using 141GB of HBM3e memory.

Distributed Training

Scale across multiple H200 GPUs with 900 GB/s NVLink interconnect.

Inference at Scale

Deploy large models with sufficient memory for extended context windows.

Ready to Deploy NVIDIA H200?

Next-generation GPU with 141GB HBM3e for large-scale AI training.