2 providers · per GPU-hour
NVIDIA GH200 cloud pricing
2 offers from 2 providers, 2 with a published on-demand price. Per GPU-hour, US$. Prices checked 28 Sept 2026.
- Lowest on-demand
- $1.99/GPU-hr
- Lowest spot
- —
- Memory
- 96 GB HBM3 (96GB); HBM3e v
Spec sheet: GH200 Grace Hopper Superchip ↗
| # | Provider | Offer | On-demand | Spot | Committed | As published (original currency) · source | Actions |
|---|---|---|---|---|---|---|---|
| 1 | Vultr | GH200 x1 | $1.99/GPU-hr | — | — | ||
| 2 | CoreWeave | NVIDIA GH200 1× GH200 Grace Hopper Superchip · 72 vCPU · 480 GB RAM | $6.50/GPU-hr | — | — |
What is the GH200 used for?
The GH200 is a data-centre GPU positioned by its maker for large-scale model training, built on the Hopper (GPU) + Grace (CPU) architecture and released in 2023. Multi-GPU servers link cards over NVLink-C2C 900 GB/s (CPU-GPU), which matters when a job is split across several GPUs.
How large a model fits on one GH200?
With 96 GB of memory, one card holds the weights of a model up to about 48 billion parameters at 16-bit precision, or about 96 billion at 8-bit. That is weights only: inference also needs room for the KV cache, and training needs several times more for gradients and optimiser state, so larger jobs are split across multiple GPUs.
| Spec (as published) | GH200 Grace Hopper Superchip |
|---|---|
| Architecture | Hopper (GPU) + Grace (CPU) |
| Form factor | superchip |
| Memory | 96 GB HBM3 (96GB); HBM3e v |
| Interconnect | NVLink-C2C 900 GB/s (CPU-GPU) |
| Released | 2023 |
| Source | maker spec sheet ↗ |
Need GH200 capacity for a longer run?
Committed and cluster pricing is often negotiated. Tell us the GPU count, duration and region.