Before you rent a GPU
Compare a complete, usable configuration. The smallest per-GPU number can belong to a large instance, a reduced-memory partition or an interruptible offer. Use this checklist alongside the sourced comparison.
- Specify the workload. Record model or renderer, software versions, memory needs, runtime and quality or latency target. Verify that the GPU and environment are supported.
- Confirm the configuration. Check GPU family, PCIe/SXM/NVL variant, memory per GPU, GPU count, CPU, host RAM and disk. “Unspecified” is a question for the provider, not an assurance of equivalence.
- Confirm region and capacity. A price list is not a reservation. Check quota, deployment availability, data location and upload time.
- Calculate the full bill. Multiply the entire instance price by billed hours, then add storage, transfer, licenses, support and tax. Check the billing minimum. Include setup and idle time.
- Read lifecycle and interruption terms. Know whether the offer is spot, what survives a stop, and how to restore checkpoints. Commitments require a separate utilization calculation.
- Record your evidence. Save the provider quote, timestamp, exact configuration and catalog source version. Run a small representative job before a longer commitment.
What does “$2 per GPU-hour” mean?
As an arithmetic example, eight GPUs at $2 per GPU-hour cost $16 per instance-hour. Ten billed hours cost $160 before additional charges. The price does not mean the provider will rent a single GPU for $2.
Reviewed September 26, 2026. Rates are snapshots, not live quotes.