Skip to content

Working comparison

Compare the options that fit this decision

Inference routing bands using the dated price examples below
RouteUse whenInline price signalPrimary unknown
Community GPUBatch can retry or queueRTX 4090 from $0.204/hrCapacity and interruption
Cloud spotCheckpointed fault-tolerant jobA100 80 GB from $0.67877/hrRegional spot availability
Managed endpointStable latency is requiredPrice depends on model and scale floorThroughput and idle minimum

Decision test

What changes the answer

  1. 01Reject GPUs that cannot hold the model and runtime overhead.
  2. 02Separate batch queue tolerance from interactive latency.
  3. 03Compare cost per successful request, not only cost per GPU hour.
  4. 04Keep a fallback route when capacity is not guaranteed.

Shareable result receipt

Fixed for this page: the same formula, assumptions and sources every visitor sees.

Use community capacity for interruptible batch work, spot capacity for fault-tolerant jobs with explicit retry, and reserved or managed endpoints when predictable latency and availability are hard requirements.

Formula

Cost per successful request = (GPU hours × hourly price + queue and retry cost + endpoint overhead) ÷ successful requests

Assumptions

  • One GPU minimum
  • Published hourly input
  • No reserved discount

Explicit unknowns

  • Measured model throughput
  • Capacity at request time
  • Retry and queue cost

Sources and freshness