🔥 A new Gorgon Halo box w/ 192GB unified memory was recently announced! Already tested on DeepSeek-V4-Flash! ACEMAGIC just revealed the new F9A powered by AMD’s Ryzen AI Max+ PRO 495 (Gorgon Halo). And this thing is basically a mini-PC local AI workstation .... stats 👇 🧠 192GB LPDDR5X-8533 unified memory 🎮 up to 160GB available to the GPU ⚡ Radeon 8065S / 40 CUs 🖥️ 16-core Zen 5 CPU 🔌 OCuLink PCIe 4.0 x4 💾 dual NVMe 🌐 dual 2.5GbE + Wi-Fi 7 🔗 dual USB4 40Gbps ACEMAGIC says it has already tested DeepSeek-V4-Flash locally, which has ... 284B total parameters 13B active/token ACEMAGIC is advertising the F9A for: 🧠 up to 120B Q4 dense models ⚡ up to 300B-class MoE models That extra 64GB moves an entire new class of huge quantized and sparse MoE models into a ~2-liter desktop. 💰Price? Probably above $5,000, but just a guess. Now I want to know one thing: How many tok/s? 😁