Home

LLM

LLM Hardware & Cost Architect

Open-source VRAM, throughput and cost calculator for LLM inference and fine-tuning.

LLM Hardware & Cost Architect

An open-source calculator that turns a model and hardware selection into concrete inference and fine-tuning numbers. A step-by-step wizard walks through model, quantization, engine, GPU and workload choices, then reports VRAM fit, tokens/sec, latency, monthly cost and TCO.

Highlights

  • Hundreds of open-source LLMs (DeepSeek, Llama, Qwen, Gemma, Mistral, Phi, Turkish models) and 30 GPU presets (NVIDIA, AMD MI300X, Apple Silicon).
  • 7 inference engines (vLLM, llama.cpp, TensorRT-LLM, SGLang, TGI, Ollama, MLX) with quant/engine/GPU compatibility rules.
  • Self-host vs. hosted-API break-even analysis and live cloud prices scraped from RunPod, Lambda and Modal.
  • Fine-tuning estimates for QLoRA, LoRA, full fine-tuning and DPO/ORPO with Unsloth acceleration.
  • Shareable scenario URLs, a scenario manager, TR/EN localization and documented methodology.

Built with React 19, Vite, Tailwind CSS v4, Express, TypeScript and PostgreSQL.

Live — https://llminferencecalc.com.trGitHub