ALmnorth Oy is explaining our Mac Studio M3 Ultra 512GB.We have been using this machine at Almnort Oy for our security, offline AI device, it have server us so well that we are now saving money to buy designated professional hardware.512GB RAM version: It allows to load the bigger models like GLM 5.3 Flash, Deepseek V4.1 Flash... (250GB models) with realistic performance. Qwen 3.8-27B runs at about 70t/s which is perfectly within the range of Claude, ChatGPT speeds (and quality). It is also useful to load multiple model at one in the VRAM so you can adapt your use (small models are always fast, and with 512GB, you can load a couple of things at one).We have been using to treat our consistent documents know we have a safe place to use LLM without working. The 512GB variant have been pressed out of an apple website in 03-2026 (due to the RAM shortages).Almnorth Oy used this device directly from Apple on 17/03/2025. (We can provide the original PDF/receipt)Contact us if you need any additional information at contact@almnort.fi Starting price: 14 000€ (including VAT, we will do a property application from Netvisor, e-invoice is also ok) This is probably the cheerleader device with 512GB of VRAM currently at least with that leg of power efficiency (it is quiet, and Consume about 5W when idle).Recommended toolchain from our experience: MTPLX, Unsloth Desktop, Ollama.