The whole card, always
No vGPU slicing and no pre-emption. The GPU is yours every hour of the month.
Bare-metal pass-throughCloud GPUs are easy to start and expensive to keep. If your workload runs for weeks, a dedicated server costs less, performs more predictably and keeps your data on hardware only you can reach.
Why it fits
No vGPU slicing and no pre-emption. The GPU is yours every hour of the month.
Bare-metal pass-throughNo per-hour meter, no egress fees and no bill that depends on how long a job ran.
Unmetered trafficDatasets and weights live on disks in a server only you can log in to. We have no agent on the machine.
Single tenantDriver, CUDA and container toolkit pre-installed on request, or a clean OS for you to build on.
Root accessRecommended servers
From single-card inference to four H100s for training.
Unlimited traffic
Unlimited traffic
Unlimited traffic
Unlimited traffic
Yes. You have root and the physical GPU. Use any framework, any container runtime and any model you are entitled to run.
No. GPU servers are available month to month. Quarterly and annual terms are discounted by 4% and 10%.
In-stock plans are provisioned automatically once payment confirms — the median time to IPMI handover is 47 minutes. Crypto payments count as confirmed after two network confirmations. Custom builds take two to five business days including a 48-hour burn-in.
From €79 a month. No setup fee, no ID, no lock-in on monthly terms.