BMT LLM Reference Released for Hardware Selection

📚 BMT LLM Reference Released for Matching Hardware to LLMs

A new resource classifies 38 modern models by deployment type, VRAM requirements, and specializations. The document provides specific recommendations: for example, for 7–12B models (Qwen 3 7B, Llama 8B), 8–12 GB of VRAM is sufficient, whereas top-tier local solutions like Llama 4 Maverick require a multi-GPU configuration with 96 GB+ VRAM. It also lists leaders among cloud models, including GPT-5.5, Claude Opus 4.8, and Gemini 3.1 Pro.

🌍 The tool creates a standardized way to match models with computing resources, which is critical for optimizing inference costs and choosing between API and self-hosted solutions.

👤 Useful for accurately selecting GPUs for specific tasks: from running lightweight models like Phi-4 Mini on mobile devices to building powerful local systems based on Llama 4.

Source 1: https://bmt-llm-reference.vercel.app