NVIDIA Hopper
NVIDIA H200 GPU Cloud & Dedicated Infrastructure
Higher-memory Hopper GPU for large models and long-context inference.
The NVIDIA H200 builds on the Hopper architecture with 141GB of HBM3e memory and significantly higher memory bandwidth than the H100. It is well suited to teams serving larger models, longer context windows, or higher-concurrency inference workloads without resorting to aggressive model parallelism.
H200 specifications
- GPU Memory
- 141GB HBM3e
- Memory Bandwidth
- Up to ~4.8 TB/s
- Architecture
- NVIDIA Hopper
- Interconnect
- NVLink / NVSwitch, PCIe Gen5
- Form Factors
- SXM
- Typical Server Configs
- 4x and 8x GPU nodes
Deployment options
- On-demand cloud GPU instances
- Reserved capacity
- Dedicated H200 servers
- Multi-node H200 clusters
Best suited for
- Serving large open-weight models (70B+) on fewer GPUs
- Long-context LLM inference workloads
- High-throughput, high-concurrency inference serving
- Training and fine-tuning workloads that are memory-bound rather than compute-bound
Less ideal for
- Small or latency-insensitive workloads where H100 capacity is more cost-effective
- Teams with strict PCIe-only server compatibility requirements
Related solutions
Managed software options
H200 frequently asked questions
When should I choose H200 over H100?
Choose H200 when your model, batch size, or context length pushes against the 80GB limit of the H100, or when memory bandwidth is your primary bottleneck. If your workloads run comfortably on H100, the incremental cost of H200 may not be justified.
Does H200 require different infrastructure than H100?
H200 is largely drop-in compatible with H100-era software stacks (drivers, CUDA, inference runtimes), though server chassis, power, and cooling requirements can differ. We handle these details as part of provisioning and managed infrastructure.
Can I reserve H200 capacity for a fixed contract term?
Yes. We can arrange reserved H200 capacity across our provider network for fixed terms, subject to current availability. Request current pricing and availability through our quote form.