[ OFFSHORE GPU ]
Rent an H100 without explaining yourself first
Large GPU clouds now attach a corporate account, an identity check and an acceptable-use committee to your training run. Here you get the whole physical card, billed by the hour from a prepaid crypto balance, with no dataset inspection and no model gating.
- Whole cards — no MIG partition, no time-slicing, full VRAM
- No usage review, no model gating, no dataset questions
- Hourly billing that stops the minute you destroy the instance
- CUDA-ready images, or start from bare Debian
[ RATE CARD ]
GPU Servers pricing
Public pricing, charged monthly or by the hour from the prepaid balance. Unmetered transfer, dual-stack addressing and mitigation are standard.
Whole cards, never MIG-sliced · no usage review · CUDA images ready · unmetered transfer
[ HOW IT WORKS ]
Why access to these cards got gated
Part of it is genuine risk management and part is liability theatre, but the effect is identical: renting a current data-centre accelerator increasingly requires a conversation about what you intend to do with it, and a permanent record of who gave the answer. For research on models a vendor would rather not host, for fine-tuning on a corpus you cannot describe to a third party, or simply for avoiding a procurement cycle, that is a bad fit.
Our Acceptable Use Policy covers crime, not capability. We do not inspect datasets, gate weights or ask what the fine-tune is for. Very few providers will hand over an H100 against a prepaid balance and no name, and that gap is most of the reason this product line exists.
See GPU plans
[ WHY NODMCAVPS ]
Sizing it without wasting money
VRAM decides almost everything. A 24 GB 4090 serves most open-weight models up to roughly thirteen billion parameters when quantised; above that you want the 80 GB on an A100 or H100. Fine-tuning needs two to three times the memory of inference on the same model, though LoRA-style methods often bring a job back within reach of a single 4090.
For rendering and simulation, a 4090 or 5090 is usually better value than a data-centre card — higher clocks, and the memory ceiling rarely binds. Whatever you rent, checkpoint to a storage volume rather than local scratch, and take the hourly billing seriously: a card left running over a forgotten weekend is the most common surprise on any GPU bill anywhere.
Full specs & options[ INCLUDED WITH EVERY PLAN ]
What you always get
[ FAQ ]
Frequently asked
Is the GPU shared with anyone else?
No. You get the entire physical card and all of its VRAM. No MIG slicing, no time-sharing, and nvidia-smi reports exactly what the specification promised.
Do you review what I train or serve?
No. There is no model gating, no dataset inspection and no use-case questionnaire. The only limits are the criminal categories in our Acceptable Use Policy.
Can I really pay with Monero for an H100?
Yes. Every product here draws on the same prepaid balance, and that balance can be funded with Monero at the same price as any other coin.
[ DEPLOY ]
Rent an H100 without in 45 seconds.
Top up, choose a region, deploy. No ID, no card, from $95/mo in crypto.