Capacity Studio
Make room for
the whole workload.
Set your task assumptions. See how they translate into waves of work, queue time and an estimated finish.
Workload assumptions
Save or load an estimate
Download your inputs and calculated result. To reopen a saved estimate, paste its JSON below; Kernveil recalculates the result from the inputs.
Estimated schedule
Browser-local96 tasks · 90 sec baseline · 8 GPUs · 75% expected utilization
- Batch completes in
- 24 min
- Final wave starts at
- 22 min
- Sustained tasks / hour
- 240
Throughput assumes a continuously full queue. A short or partial batch can have a lower realized rate.
- Effective time / task
- 2 min
- Waves
- 12
Your queue, wave by wave
8 tasks per full wave
The assumptions behind the estimate
- Identical GPUs are available together at the start.
- Each independent task uses one GPU for the same effective duration.
- A partial final wave still takes a full task duration.
Expected utilization scales task duration. It is your input, not a measurement. Transfers, memory limits, dependencies, retries and hardware differences are outside this model.
Read the calculation model