🧮 LLM GPU Planner

Two tools to help you predict latency and estimate GPU requirements. Use the tabs below 👇

Estimate GPUs needed from max concurrency.

Formula: GPUs = ceil(max_concurrency × target_stream_tps / (per_gpu_tps × util_cap))

Calls per month

Average call duration (minutes)

Max concurrent active users

Target stream tokens/s per call

Utilization cap (headroom)

0.1 0.95

Per-GPU tokens/s (measured or expected)

Optional: derive Per-GPU TPS from a measured cluster

Measured cluster tokens/s

#GPUs in that measurement