Rent GPU servers on-demand
Your own Ubuntu VM with the whole GPU passed through and root SSH access. NVIDIA RTX 4090, A100, H100 and more, billed per second with no contracts and no upfront costs.
GPU server rental prices
Live rates per GPU-hour. Each GPU is passed through whole to your VM, never shared.
How GPU rental works on OpenRelay
From signup to SSH in three steps.
Pick a GPU
Choose one of 6 GPU models in the table above, from $0.18/hr. Its Rent link carries your pick through signup to the VM form.
Configure your VM
Choose how many GPUs, add your SSH key and deploy. The VM boots Ubuntu with the GPUs passed through to it.
Connect over SSH
Log in with root access and run your workload. Billing is per second, with no minimum and no egress fees.
GPU rental pricing vs competitors
On-demand rates per GPU-hour.
| Provider | RTX 4090 | H100 | ||
|---|---|---|---|---|
| OpenRelay | $0.35 | $2.60 | Free | Per-second |
| RunPod | $0.74 | $3.49 | Free | Per-minute |
| Lambda Labs | N/A | $2.49 | Free | Per-hour |
| AWS | N/A | $4.76 | $0.09/GB | Per-second |
| GCP | N/A | $4.15 | $0.12/GB | Per-second |
Why OpenRelay is cheaper than AWS, GCP, and RunPod
Up to 76% less than hyperscalers: here's why the math works out.
Distributed infrastructure
We aggregate GPU capacity from providers worldwide instead of building our own data centers. This eliminates the massive capital expenditure that hyperscalers pass on to you as markups.
No egress fees
AWS, GCP, and Azure charge $0.08, $0.12 per GB for data transfer out. On large inference workloads, egress can add 20-40% to your bill. OpenRelay charges $0 for egress, ever.
Consumer GPU efficiency
RTX 4090s deliver 82.6 TFLOPS of FP32 compute at $0.35/hr. An A10G on AWS costs $1.21/hr for less throughput. For inference workloads under 34B parameters, consumer GPUs win on price-performance.
Why rent GPUs instead of buying?
Renting gives you flexibility, lower risk, and access to the latest hardware.
No upfront capital
A single RTX 4090 costs $1,600+. An H100 costs $30,000+. Renting lets you access the same hardware with zero upfront cost and no depreciation risk.
Scale up and down instantly
Need 10 GPUs for a training run and 2 for inference? Scale your GPU fleet in minutes instead of waiting weeks for hardware procurement.
No maintenance burden
Hardware fails. Drivers need updates. Cooling systems break. When you rent GPUs, the provider handles all of this. You just run your workloads.
Access the latest hardware
New GPU generations drop every 1-2 years. Renting means you move to the new one when it ships instead of being stuck with last-gen hardware.
Try before you commit
Not sure which GPU fits your workload? Rent two for an hour, benchmark, and keep the better one.
No power or cooling costs
A single H100 draws 700W. Running 8 GPUs 24/7 costs $800+/month in electricity alone. Renting includes all power and cooling in the hourly price.
Frequently asked questions
How do I rent a GPU server?
Pick a GPU in the table on this page and press Rent. Sign up, choose how many GPUs, add your SSH key and deploy: you get an Ubuntu VM with the GPU passed through and root SSH access.
What is the minimum rental period?
There is no minimum rental period on OpenRelay. You can rent a GPU for as little as one second. Billing is per-second with no minimum commitment, no setup fees, and no cancellation fees. Spin up a GPU, run your workload, and shut it down when you're done.
Can I rent multiple GPUs at once?
Yes. Put several GPUs in one VM where a node has them free, or run several VMs. The VM form shows which GPU counts each model can take right now.
What kind of GPU access do I get?
You get full, dedicated GPU access via VFIO passthrough. The GPU is not shared or virtualized: your VM gets bare-metal performance with direct access to all CUDA cores, VRAM, and tensor cores. You also get full root SSH access to your server.
Is renting a GPU cheaper than buying?
For most use cases, yes. An RTX 4090 costs $1,600+ to buy, plus electricity, cooling, and maintenance. At $0.35/hr on OpenRelay, you'd need to run it 24/7 for over 6 months before buying becomes cheaper: and that doesn't include power costs, maintenance, or the opportunity cost of capital. For variable or part-time workloads, renting is significantly cheaper.
What is the cheapest GPU server available?
OpenRelay's cheapest GPU server is the RTX 3090 with 24GB GDDR6X at $0.18/hr. All prices include storage and networking with no egress fees.
Do you offer long-term or volume GPU pricing?
Yes. Contact us for custom pricing on long-term or larger deployments.
Rent your first GPU server
No contracts. No egress fees. Just fast, affordable GPU compute.