Skip to main content

Running Benchmarks

Emmy lets you run standardized benchmarks across different GPU types and model configurations, so you can find the best performance-to-cost ratio before committing to a deployment.

Metrics​

Each benchmark captures:

  • Throughput — tokens per second (TPS)
  • Time to first token (TTFT)
  • Inter-token latency (ITL)
  • GPU memory utilization
  • Cost per 1M tokens (based on CloudRift GPU pricing)

Running a benchmark​

Select a recipe or custom configuration, choose one or more GPU types, and launch. Results are saved automatically and can be shared as experiments.