Maximum Memory for AI Workloads
141GB HBM3e memory—1.9x more than H100. Run larger models, longer contexts, and bigger batches without memory constraints.
Memory Where You Need It
H200 removes memory bottlenecks from your AI pipeline.
141GB HBM3e Memory
1.9x more memory than H100. Fit larger models and longer context lengths without memory constraints.
4.8 TB/s Bandwidth
1.4x higher memory bandwidth than H100 for faster data movement and improved throughput.
NVLink 4.0
900 GB/s GPU-to-GPU bandwidth enables efficient scaling across multi-GPU configurations.
Hopper Architecture
Same proven Hopper architecture as H100, with expanded memory for larger workloads.
Pre-configured Environment
PyTorch, TensorFlow, CUDA 12, and popular ML frameworks ready to use immediately.
Expert Support
Our ML infrastructure team helps with environment setup, optimization, and debugging.
Technical Specifications
H200 SXM specifications for reference.
| Specification | Value |
|---|---|
| GPU Memory | 141GB HBM3e |
| Memory Bandwidth | 4.8 TB/s |
| FP8 Performance | 3,958 TFLOPS |
| FP16 Performance | 1,979 TFLOPS |
| FP32 Performance | 989 TFLOPS |
| NVLink Bandwidth | 900 GB/s |
| TDP | 700W |
| Form Factor | SXM5 |
What H200 Excels At
Memory-intensive workloads where H100 isn't enough.
Large Model Inference
Deploy 70B+ parameter models for production inference. 141GB memory fits models that require multi-GPU on H100.
Memory-Intensive Training
Train models with large batch sizes and long context lengths without gradient checkpointing overhead.
RAG Applications
Retrieval-augmented generation with large context windows benefits from H200's expanded memory.
Scientific Computing
HPC workloads with large datasets that benefit from high memory capacity and bandwidth.
On-Demand & Reserved Pricing
Flexible pricing for memory-intensive workloads.
1x H200 SXM
- 1× NVIDIA H200 SXM
- 141GB HBM3e
- 24 vCPU
- 240GB RAM
- 1TB NVMe
- Pre-configured ML Environment
4x H200 SXM
- 4× NVIDIA H200 SXM
- 564GB HBM3e
- 96 vCPU
- 960GB RAM
- 4TB NVMe
- Pre-configured ML Environment
8x H200 SXM
- 8× NVIDIA H200 SXM
- 1.1TB HBM3e
- 192 vCPU
- 1.9TB RAM
- 8TB NVMe
- Pre-configured ML Environment
Need a custom configuration? Contact us for a quote.
Frequently Asked Questions
How does H200 compare to H100?
H200 has 1.9x more memory (141GB vs 80GB) and 1.4x higher memory bandwidth (4.8 TB/s vs 3.35 TB/s). Compute performance is the same. Choose H200 when you need more memory capacity or are memory-bound.
When should I choose H200 over H100?
Choose H200 if you're running large model inference (70B+ parameters), training with large batch sizes, or working with long context lengths. If compute is your bottleneck, H100 offers better price/performance.
What frameworks are supported?
All major frameworks: PyTorch, TensorFlow, JAX, Hugging Face Transformers. H200 uses the same CUDA/cuDNN stack as H100, so existing code works without modification.
Is reserved pricing available?
Yes. Reserved instances offer 20-40% discounts compared to on-demand pricing. Contact our team for a quote based on your commitment term.
Can I mix H200 with H100 in a cluster?
For most workloads, we recommend homogeneous clusters (all H200 or all H100). Mixed configurations can work for certain pipelines—contact our team to discuss your architecture.
Get Your H200 Instance Today
Talk to our team about your large-scale AI workload. We'll help you decide between H200 and H100.