NVIDIA GPU VPS, CUDA-ready in minutes
Monthly GPU servers on NVIDIA A16 and A40 hardware: from a $90 fractional slice for inference and encoding up to a dedicated 4-GPU box with 64 GB of GPU memory. Ten locations worldwide, full root access, payable by PayPal or 18 cryptocurrencies.
Rent a slice of the card, or the whole thing
1/24 to 1/2 of a GPU
A dedicated share of the card with its own slice of GPU memory (2 to 8 GB). CUDA works exactly as on a full card; you just size the memory to your model or workload.
A full A16 with 16 GB
The whole card to yourself from $720 per month, with 6 vCPU, 64 GB RAM and NVMe storage sized to match.
2x and 4x A16 boxes
Scale to 32 or 64 GB of total GPU memory across multiple cards for parallel inference workers or batch pipelines.
Every GPU plan we sell
| GPU | GPU memory | vCPU | RAM | NVMe | Transfer | From |
|---|---|---|---|---|---|---|
| A16, 1/8 GPU | 2 GB | 2 | 8 GB | 50 GB | 1 TB | $90/mo |
| A40, 1/24 GPU | 2 GB | 1 | 5 GB | 90 GB | 3 TB | $120/mo |
| A16, 1/4 GPU | 4 GB | 2 | 16 GB | 80 GB | 2 TB | $180/mo |
| A16, 1/2 GPU | 8 GB | 3 | 32 GB | 170 GB | 3 TB | $360/mo |
| A16, full GPU | 16 GB | 6 | 64 GB | 350 GB | 6 TB | $720/mo |
| 2x A16 | 32 GB | 12 | 128 GB | 700 GB | 10 TB | $1,440/mo |
| 4x A16 | 64 GB | 24 | 256 GB | 1.2 TB | $2,873.75/mo |
Our retail pricing in USD at the lowest-priced location; some metros carry a surcharge and not every plan is offered in every location. The configurator shows live pricing and availability for your chosen city.
Honest sizing: A16 and A40, not H100
Runs well here
Inference on small and mid-size models, image generation on the 8 to 16 GB tiers, video encoding and transcoding, virtual desktops and cloud gaming, CUDA development and CI runners.
Not the right tool
Training or serving frontier-scale LLMs. An A16 is not an H100, and we would rather say so here than after you have deployed. If your job needs 80 GB of HBM, rent that class of card elsewhere and keep us for everything around it.
Why monthly GPU
A flat monthly price from your prepaid balance instead of a metered hourly bill. Top up by PayPal or crypto, deploy, and know what the month costs before it starts.
GPU VPS questions
Which GPU models do you offer?
NVIDIA A16 (fractional slices up to 4x full cards) and an entry NVIDIA A40 slice. We do not sell A100 or H100 instances, and we will not pretend an A16 substitutes for one.
Can I run or train large language models on these?
Inference on small and mid-size models fits the 8 to 64 GB tiers well. Fine-tuning small models is possible on the multi-GPU boxes. Large-scale training belongs on data-center cards with more memory bandwidth than this class offers.
Do the images come with drivers and CUDA?
Yes. GPU plans deploy CUDA-ready Linux images with NVIDIA drivers matched to the plan family, so nvidia-smi works on first boot.
How does billing work?
Plans are monthly and paid from your prepaid account balance. Top up with PayPal or any of 18 cryptocurrencies including Bitcoin and Monero; no card or ID required.
Is every plan available in every location?
No. GPU capacity varies by city across our ten locations, and some metros price slightly higher. The order form shows live availability and the exact price for your chosen location.
Deploy a CUDA-ready GPU server
From a $90 slice to a 4-GPU box. Ten locations, monthly pricing, crypto welcome.
Browse GPU plans