Skip to content

Case studies

What a Traction Runtime case study will contain: workload, configuration, measured method and results, published only with the customer's written approval.

Book a Runtime AssessmentThe benchmark harness

  • Workload

    [Workload]

  • Configuration

    [Configuration]

  • Measured method and results

    model, hardware, driver/CUDA, container digest, trace, concurrency sweep, the four configurations, three runs, goodput at SLO, p95 TTFT/TPOT, cost per million tokens, and the reproducing command

  • Published only with the customer's written approval

    [Approval record]

The benchmark harness
Book a Runtime Assessment