AI & LLM Calculators
Free, interactive calculators for planning LLM systems — sizing, cost, and speed. Each is built on the underlying architecture math (not a stale price sheet), so the answers stay accurate. No sign-up.
LLM VRAM & GPU Sizing Calculator
Does the model fit, and how many GPUs?
Model weights + KV cache + overhead → total VRAM and the number of H100/H200/A100 GPUs, with a tensor-parallelism flag when weights exceed one card.
Open →$LLM Inference Cost Calculator
What does it cost to run, per month?
Cost per request and per month from token usage and price, the blended $/1M-token rate, prompt-cache savings, and the volume where self-hosting beats the API.
Open →⚡LLM Throughput & Latency Calculator
How fast will it serve?
Tokens per second, time-to-first-token, and batched aggregate throughput on a memory-bandwidth roofline — and whether you are bandwidth- or compute-bound.
Open →How the three fit together
Planning to serve an LLM is three questions in sequence, and each calculator answers one. First, does it fit — the VRAM & GPU sizing calculator turns a model, precision, context, and concurrency into the VRAM and GPU count you need. Then, how fast — the throughput & latency calculator gives tokens per second and time-to-first-token on a memory-bandwidth roofline. Finally, what it costs — the inference cost calculator estimates the monthly bill and the break-even where self-hosting beats a per-token API.
For the architecture behind the numbers, see why HBM bandwidth is the real inference bottleneck, AI infrastructure sizing for LLM workloads, and GPU utilization and inference economics.