Infer18Build your cluster
Engineering estimateAll sizing is an estimate until validated on silicon.
Back to the storyBack
Step 01 of 07

What do you need to serve?

Set your volume and conversation load. Pick a preset or drag the sliders.

Tokens generated per month2T
Peak conversations at once20,000
Typical conversation length
Main use case
Stage 01 of 06ChipPurpose-built inference silicon. Your model is part of the hardware, so there is nothing to load.
Power~150–200 WIn build752
Parts0 / 0
RecommendedYour build includes 1 spare node, one set per model slice, so failed nodes swap for the same slice.
Your build so far: 6 racks, 47 nodes, 295 kW