The whole card, passed through.
Not a slice, not a time-share, not MIG. The GPU appears on your PCIe bus and nothing else on earth is scheduled on it.
GPU servers, billed by the month, not the hour
Whole cards, passed through to your machine. No queue, no scheduler, no neighbour, and no meter running while you think.
- Jurisdiction
- Moldova · non-EU
- Card
- Yours, passed through
- Ready in
- ~15 minutes
- From
- $229/month
Size your server
RTX 6000 Ada
- CPU16c / 32t
- Memory128 GB ECC
- NVMe2 × 2 TB
- Network1 Gbps
3 builds
Every GPU build, billed by the month
| Plan | GPU | CPU | Memory | Storage | Network | Ready in | Price | Order |
|---|---|---|---|---|---|---|---|---|
| RTX 4000 Ada Inference, rendering, fine-tuning small models | RTX 4000 Ada · 20 GB | Ryzen 7 7700 · 8c/16t | 64 GB DDR5 | 2 × 1 TB Gen4 NVMe | 1 Gbps unmetered | ~15 min | $229$183/mo | Configure RTX 4000 Ada |
| RTX 6000 Ada 48 GB models, video pipelines | RTX 6000 Ada · 48 GB | EPYC 7302P · 16c/32t | 128 GB DDR4 ECC | 2 × 2 TB Gen4 NVMe | 1 Gbps unmetered | ~15 min | $549$439/mo | Configure RTX 6000 Ada |
| L40S Training, batch inference at volume | L40S · 48 GB | EPYC 7443P · 24c/48t | 256 GB DDR4 ECC | 2 × 3.84 TB Gen4 NVMe | 10 Gbps unmetered | ~15 min | $749$599/mo | Configure L40S |
Every build: whole GPU passed through · IPMI · 1 IPv4 · routed IPv6 /64 · always-on DDoS filtering · unmetered port · full API. Multi-card chassis are quoted per machine — ask, we build them.
The hourly trap
A card at $1.20 an hour costs $876 a month. Ours does not.
Hourly GPU pricing is designed for bursts, and almost nobody bursts. If the card is busy more than a third of the time, a monthly machine is cheaper — and it stops charging you for the hours you spend reading logs.
No queue and no preemption.
Your job starts when you start it and finishes when it finishes. Nothing evicts you because somebody with a bigger budget arrived.
48 GB where it matters.
The RTX 6000 Ada and the L40S carry 48 GB each — enough to hold a model that would otherwise need sharding, which usually costs more than the card.
Drivers are yours to choose.
We hand over bare metal and an ISO of your choosing. CUDA version, kernel, container runtime — pinned by you, not by a platform that upgrades on its own schedule.
No upsells
In the price of every GPU server
No protected-IP tier, no per-snapshot charge, no support plan to buy.
The whole machine
One tenant. Nothing else boots on it, nothing else touches the disks, nothing else shares the memory controller.
IPMI, not a support ticket
Full out-of-band control: power, console, virtual media. Reinstall at three in the morning without asking anyone.
IPv4 + a routed /64
One IPv4 included, a /29 on request, and a whole /64 of IPv6. Reverse DNS is yours to set.
DDoS filtering
Layer 3/4 scrubbing upstream of the port, on by default, at no charge and with nothing to enable.
Reinstall or bring an ISO
Twenty prepared images, or mount your own and watch it boot over the console. Nothing survives a wipe.
API and SSH keys
Order, reinstall, reboot and destroy from your own code. Keys installed at handover, password login off.
Before you order
What people ask before renting a GPU server
Anything else, support answers in minutes — and says no plainly when the answer is no.
Open a ticketWhich card should I take?
For inference and rendering under 20 GB, the RTX 4000 Ada. For 48 GB models or heavy video, the RTX 6000 Ada. For sustained training and batch inference, the L40S — more memory bandwidth and a chassis with the CPU and NVMe to keep it fed.
Can I have several cards in one machine?
Yes, quoted per chassis. Tell support how many and which, and whether you need NVLink — the answer to that last one is usually no, and we will say so rather than sell you the more expensive build.
Is it really fifteen minutes?
Yes. The cards sit in racked machines that are burned in and waiting; ordering triggers a wipe and a reinstall. What takes the time is the memory training and the POST on a chassis this size, not somebody walking to a rack.
Do you allow model training on any data?
The same two exceptions as everywhere else on this site apply, and nothing more — see how complaints are handled. We do not inspect what runs on your machine and we have no interest in doing so.
The rest of the range
Compare the GPU line against the other bare-metal lines
Same jurisdiction, same network, same refusal to ask who you are. They differ on the one thing named in each title.
Ready when you are
Pick a card. Pay in crypto. Hold root in a minute.
No account needed to see every price, no identity needed to order. $229 a month gets you a server nobody can talk us into switching off.
- 72-hour refund on a first order
- Free migration from your current host
- No setup fee, no minimum term