ModelFitRun fit check→
Home/Calculators
Four auditable tools

LLM GPU
calculators.

Start with the narrowest question you can answer reliably. Each tool exposes its formula, omits unknowns instead of inventing them, and links back to the full deployment worksheet.

01
Full deployment worksheet

LLM VRAM calculator

Combine model weights, context, concurrent sequences, KV precision, runtime workspace, GPU count and operating headroom.

  • Total VRAM
  • GPU fit
  • vLLM command
Open calculator ↗02
Architecture-aware cache

KV cache calculator

Isolate how context length, active sequences, cache precision and KV-head sharding change serving memory.

  • Context
  • Concurrency
  • FP8 cache
Open calculator ↗03
Topology planning

Multi-GPU calculator

Estimate the smallest 1×, 2×, 4× or 8× topology that fits a selected model and GPU memory profile.

  • Tensor parallel
  • GPU count
  • 85% ceiling
Open calculator ↗04
Unknown checkpoints

Model weight memory calculator

Estimate the weight-loading floor for a custom parameter count and precision when the model is not yet in the directory.

  • Parameters
  • Quantization
  • Weight floor
Open calculator ↗
One calculation is not a benchmark

Validate the chosen envelope.

Use the tools to narrow the topology, then test the exact runtime, checkpoint and GPU before committing capacity.

Read the methodologyCompare GPU profiles