Inference calculator
VRAM, fit, throughput and parallelism for one model on real GPUs.
-
01
Inference calculator — overview
What the calculator answers, what it needs from you and how to read the cards it returns.
4 min read · updated 2026-10-26 -
02
The input deck and Advanced: batch, targets, LoRA, speculative decoding
What the input deck already decides, and what batch, latency targets, LoRA adapters and speculative decoding change when you open Advanced.
5 min read · updated 2026-10-26 -
03
How the calculator picks TP / PP / EP
The order the calculator tries tensor, pipeline and expert parallelism, and why it avoids pipeline stages when TP is enough.
4 min read · updated 2026-10-26 -
04
Multi-node, rack-scale and recommendation order
Why a single card that fits always ranks first, when rack-scale systems appear and how multi-node setups are ordered.
3 min read · updated 2026-10-26