GPU instances

Lambda vs RunPod

Choose Lambda when you want a fixed price per GPU-hour on a named instance. Choose RunPod when community-cloud or serverless pricing can undercut that, and you can live with a rate that changes by card and cloud type.

Lambda

On-demand NVIDIA instances with a published hourly rate

From $0.69/hr

Strengths

  • Each instance type has a posted price per GPU-hour
  • On-demand instances bill in one-minute increments

Limits

  • The $0.69 plan is a Quadro RTX 6000, not a current datacenter GPU
  • An 8-GPU H100 SXM is $3.99 per GPU-hour

Vendor pricing Alternatives

RunPod

GPU pods and serverless inference, priced by the hour

Hourly, by GPU

Strengths

  • Community Cloud is usually cheaper than a single-tenant GPU
  • Serverless workers fit bursty inference

Limits

  • Community Cloud shares machines and the rate moves with the card
  • Secure Cloud costs more than Community Cloud for the same GPU

Vendor pricing Alternatives