NVIDIA Tesla T4 mieten.

A pre-configured VM that pairs the NVIDIA Tesla T4 with a fixed amount of vCPU, RAM, and SSD on a hyperscaler (AWS, Google Cloud, Azure, Oracle, Vultr, etc.). You pay one hourly rate for the whole bundle; the per-GPU rate on the table is just the bundle price divided by the GPU count. Like raw GPU rental it is on-demand — billed while the instance runs, no subscription — though hyperscalers also sell reserved or committed-use plans that trade a term commitment for a lower rate.

Hyperscaler bundles include managed networking, premium NVMe storage, an SLA, and 24/7 support. P2P marketplaces (Vast.ai, RunPod community, io.net) skip those line items, which is why their raw GPU rates are often 3–10× cheaper. The trade-off is reliability and integration: a hyperscaler instance plugs into your existing VPC, IAM, and observability stack out of the box.

Sort the table by the $/hr per GPU column — the lowest row is the cheapest hyperscaler today. Pricing shifts as providers re-price spot capacity and refresh quotas, so the leader can change week to week. Click Launch on a row to head to that provider's sign-up page (affiliate link, doesn't change the rate you pay).

Yes. All listed hyperscalers expose standard compute APIs — bring your own AMI/image, mount custom volumes, install your own drivers, run any container runtime you like. The bundle just sets the hardware shape; the OS layer is yours to configure.

If you can tolerate a P2P marketplace, the Overview tab's "Where to rent it" table usually has the lowest hourly rate by a wide margin. Use a hyperscaler bundle only when you need managed networking, SLAs, or are already inside that cloud's ecosystem (compliance, data-egress costs, IAM).