Arm, not x86
The Grace CPU uses the Arm instruction set. Docker images and Python wheels built only for x86-64 won't run, so use arm64 builds of your inference server and check every dependency before you move a workload.
GPU rental prices · Grace Hopper
NVIDIA GH200 rental prices
NVIDIA GH200 rental starts at $1.99 per GPU-hour (Vultr, Superchip, bare metal) across the 2 offers we track from 2 providers. The typical price is $2.14, and one GPU running all month costs about $1,453 at the lowest rate. Each GH200 pairs a Hopper GPU with a 72-core Arm Grace CPU on one module. Per GPU-hour it rents for less than an H100 SXM at the same provider, but your software has to run on Arm.
- Lowest per GPU-hour
- $1.99 Vultr
- Typical per GPU-hour
- $2.14 median, USD
- One GPU all month
- $1,453 at the lowest rate
- Providers
- 2 2 offers
On-demand prices
NVIDIA GH200 price per GPU-hour
Every GH200 offer in our catalog, cheapest first. Multi-GPU machines are divided by their GPU count, so you can compare one GPU with one GPU. Monthly cost assumes the GPU runs all 730 hours.
| Provider and offer | Version | GPU memory | GPUs | Per GPU-hour | Per month, 1 GPU | Checked |
|---|---|---|---|---|---|---|
| VultrVultr NVIDIA GH200 · 1 GPU | Superchip, bare metal | 96 GB | 1 | $1.99 | $1,453 | 3 Nov 2026 |
| LambdaLambda NVIDIA GH200 · 1 GPU | Superchip | 96 GB | 1 | $2.29 | $1,672 | 3 Nov 2026 |
On-demand list prices before tax, from each provider's pricing page. Community and reserved prices, where a provider offers them, are lower; see the notes below. Full offer details, storage and billing terms.
Before you rent
What changes the price
Cheaper than an H100 at the same provider
Lambda lists the GH200 at $2.29 per GPU-hour and the H100 SXM at $4.29. Vultr rents the GH200 as a bare-metal server at $1.99 an hour.
Looking for a purchase price?
NVIDIA doesn't list a purchase price on its GH200 page. This page covers renting one by the hour.
AI models
Which models fit on the GH200
Memory to load each model size on 96 GB GPUs, with a 20% allowance for runtime buffers, and what that many GPUs cost per hour at the lowest rate above. Long contexts and many parallel users need more memory. More than one GPU only works if your inference server splits the model across them.
| Model size | 4-bit | 8-bit | 16-bit |
|---|---|---|---|
| 8BLlama 3.1 8B, Qwen3 8B | 4.5 GB1 GPU · $1.99/hr | 9.0 GB1 GPU · $1.99/hr | 17.9 GB1 GPU · $1.99/hr |
| 32BQwen3 32B, Gemma 4 31B | 17.9 GB1 GPU · $1.99/hr | 35.8 GB1 GPU · $1.99/hr | 71.6 GB1 GPU · $1.99/hr |
| 70BLlama 3.3 70B, Qwen 72B | 39.2 GB1 GPU · $1.99/hr | 78.3 GB1 GPU · $1.99/hr | 156.5 GB2 GPUs · $3.98/hr |
| 120Bgpt-oss-120b class | 67.1 GB1 GPU · $1.99/hr | 134.2 GB2 GPUs · $3.98/hr | 268.3 GB3 GPUs · $5.97/hr |
| 685BDeepSeek-V3.2 | 382.8 GB4 GPUs · $7.96/hr | 765.6 GB8 GPUs · $15.92/hr | 1,531.1 GB16 GPUs · more than one 8-GPU server |
Weights only, from the parameter count. Check a specific model, context length and user count in the model hardware calculator.
Hardware
NVIDIA GH200 specifications
| Specification | GH200 |
|---|---|
| GPU | Hopper, 96 GB in the offers listed here |
| CPU | 72 Arm cores (Grace) |
| CPU–GPU link | NVLink-C2C, 900 GB/s, memory-coherent |
| Combined fast memory | Up to 624 GB of CPU and GPU memory in NVIDIA's largest configuration |
From NVIDIA's product pages. Tensor figures marked “sparse” assume structured sparsity; dense throughput is half.
Questions
GH200 rental FAQ
How much does it cost to rent a GH200?
$1.99 per hour is the lowest on-demand price we track, from Vultr. The typical price is $2.14, and one GH200 running all month costs about $1,453 at the lowest rate.
Can a GH200 run a 70B model?
Yes. At 8-bit the weights need about 78.3 GB with a 20% allowance, which fits in the 96 GB of GPU memory offered here with about 18 GB left for the conversation cache. At 4-bit they need 39.2 GB.
Compare
Other GPU rental prices
Prices and availability change, and we don't check stock. Confirm the current price and region with the provider before you rent. Sources: www.nvidia.com, lambda.ai, www.vultr.com . See our affiliate disclosure.