NVIDIA Data Center · Hopper

Rent NVIDIA H200 GPU compute by the hour

NVIDIA H200 expands Hopper with 141 GB of HBM3e memory, making it relevant to memory-intensive AI inference and training. Check GPU-Link for live provider listings rather than relying on static inventory claims.

Memory141 GB HBM3e
GenerationHopper
Best fitMemory-heavy AI

When H200 memory can help

Large models and long-context workloads can be constrained by GPU memory before they are constrained by raw compute. The H200’s larger high-bandwidth memory capacity can reduce some of those limits for workloads designed to use it.

H200 versus smaller accelerators

A larger accelerator is not automatically the best value. If your model fits comfortably on a smaller GPU, a lower-cost listing may be more efficient. Choose based on measured workload needs, not model name alone.

Provider-set pricing

On GPU-Link, providers control their own hourly rates. The number of active H200 listings can change, so use the marketplace to confirm exact hardware, readiness and current price.

Frequently asked questions

How much memory does NVIDIA H200 have?

NVIDIA H200 is commonly specified with 141 GB of HBM3e memory.

Is H200 useful for LLM inference?

Its large high-bandwidth memory can be useful for memory-heavy inference, but suitability depends on the model and serving configuration.

Where do I see a live H200 rental price?

Use the GPU-Link marketplace. Prices are set by individual providers.