Deterministic by design
The same inputs always give the same answer. There is no random telemetry, no hidden model guessing at your utilization, and no quietly changing numbers between visits.
About
ClusterSim started as a way to answer a question that never has an honest answer on a pricing page: what would this actually cost, and how do you know? It sizes GPU clusters, runs your training, inference and AI agent workloads against them, and compares renting against owning — with every assumption left visible so you can argue with it.
Navin Mishra
Questions, corrections and price sources are all welcome — they make the estimates better. If you spot a number that looks wrong, say so and it gets checked.
Six rules the whole app is built on. They are the reason ClusterSim sometimes refuses to give you a number.
The same inputs always give the same answer. There is no random telemetry, no hidden model guessing at your utilization, and no quietly changing numbers between visits.
A cost only appears when a price record carries a source and a date, or when you enter the rate yourself. ClusterSim never fills a gap with an invented number.
When pricing, capacity or model details are absent, the result is labelled incomplete or cannot be estimated — shown as unavailable rather than as zero.
Results are marked Calculated, Illustrative, Incomplete or Cannot estimate, so you can tell how much of an answer rests on real inputs.
The planner runs in your browser and works without an account. Sign in only if you want your workspace saved across devices; assistants get read-only access you approve.
Optional AI write-ups restate numbers the engine already computed. They cannot introduce new costs, specs or benchmarks, and are available to signed-in users only.
Describe what you need in plain English, review what ClusterSim understood — stated, assumed or missing — then compare the clusters it recommends.