Up to 64% less than the big clouds
A GPU cloud with no egress fees, no contracts and no sales call. Rent GPUs by the hour, from an RTX 4090 up to a B200.
- H100 from, per hour
- -----
- GPU VMs listed
- ---
- GPU containers listed
- --
- Countries
- -
Live supply
The best GPU deals right now
Consumer and workstation
Data center
Move your data for free
$0
Per GB, in or out of your GPU. You pay for GPU time, not for traffic.
What the big clouds charge for data out
AWS$0.09/GB
Google Cloud$0.12/GiB
Azure$0.087/GB
The same GPU for less per hour
| Provider | 1× H100 80 GB, per hour | 1× A100 80 GB, per hour |
|---|---|---|
 Varity | $4.41 | $2.19 |
| AWS | $6.881.5× | $3.431.5× |
| Google Cloud | $11.062.5× | $5.072.3× |
| Azure | $12.292.7× | $3.671.6× |
On-demand rates per GPU-hour, before storage. Varity’s is the lowest live rate.
Privacy
Only your key opens the machine
- Your private key stays on your computer
- The Portal takes only the public key
- Varity puts your key, and no other, on the machine
Your GPU runs in our supplier’s Tier 3 and Tier 4 data centers, which meet GDPR, ISO 27001 and SOC 2 standards
What you can run on the GPU cloud
PyTorchTrain and fine-tune
Leave room in VRAM for weights, optimizer state and batches. For multi-GPU runs, check the interconnect on the profile you pick.
Hugging FaceServe your own models
Bring a compatible model and inference server, then test latency and concurrency on the GPU you chose
BlenderRender and simulate
Match the GPU to your render engine, with enough host memory and storage for scenes and results
GPU cloud questions, answered
What is a cloud GPU server?
A GPU in a data center that you rent and reach over the internet. On Varity you choose a GPU container, a GPU virtual machine or bare metal, add your SSH key and pay by the hour.
Which GPUs can I rent?
The Developer Portal lists every available GPU with its exact hourly rate before you deploy, from an RTX 4090 up to data center GPUs such as the H100 and B200. Availability follows live supply.
How is GPU time billed?
GPU compute has a 2-hour minimum, billed at deploy. After that it is billed hourly, and you can delete it any time. There are no egress fees.
Who can access my GPU machine?
Only your key opens it. Your private key stays on your computer, the Portal takes only the public key, and Varity puts your key, and no other, on the machine.
Do I need a GPU to use AI models?
No. If you only need an API for AI models, the Varity AI Gateway gives you one OpenAI-compatible API for its live model catalog.
Find your GPU in the Portal
See every GPU and its exact hourly rate in the Developer Portal before you deploy