Find GPU.
Browse regional inventory, filter by model and VRAM, then create an instance. Minute-level start and per-second metering where available.
0 configurations · 0 of 0 GPUs available
The GPU catalog exposes model, VRAM, price, availability, data center, manufacturer, max GPUs per instance, and CUDA versions. It does not bind exact vCPU/system RAM per SKU; create uses minima minVcpuCount=2, minMemoryInGb=15. Container disk default 20GB; volume default 20GB (adjustable). Storage ~$0.10/GB/month while running.
Listed price is the per-GPU hourly rate (USD/hour). Standard and Shared pools appear as separate rows. Disk is billed separately.
Host lines like "188GB RAM · 12 vCPU" are machine-level specs not exposed by the public catalog API. We only show verifiable catalog fields and create minima — never invented machine CPU/RAM.
Environment · ready to run.
- PyTorch 2.6 + CUDA 12.8
- vLLM inference image
- ComfyUI workflow image
- JupyterLab
- NVLink / InfiniBand
- Shared storage from 100GB
- Public port mapping
- VPC isolation
- Per-second billing
- Snapshots & clones
- GPU utilization monitoring
- Auto-stop protection