🔥 GMKtec isn't getting the 192GB Local AI market to itself. ... but this has a few interesting differences ... CHUWI unveiled the UniBox AI495 Pro at IFA. And they stuffed a ridiculous amount of AI hardware into 2.9 liters Stats 👇 🧠 Ryzen AI Max+ PRO 495 ⚙️ 16 Zen 5 cores / 32 threads 🎮 Radeon 8065S — 40 CUs 💾 192GB LPDDR5X-8533 unified memory 🎮 Up to 160GB allocated as VRAM ⚡ 273 GB/s memory bandwidth 💽 3× M.2 PCIe SSD slots 🌐 Dual 10GbE 🔌 OCuLink ⚡ 2× USB4 📡 Wi-Fi 7 CHUWI says it can run 300B-parameter models locally at 4-bit. (Same as GMKtec) ⚠️ That's a capacity claim, and really no 300B TPS benchmark shown yet. 👀 But OCuLink makes this one especially interesting. You could attach an NVIDIA GPU when you want CUDA... OR potentially attach an AMD Radeon with 24/32/48GB VRAM and use a multi-GPU runtime to split a huge model across the 8065S + discrete Radeon. A 48GB card could theoretically mean roughly 160GB + 48GB = 208GB of aggregate GPU model capacity. 🤯 ⚠️ Not one unified VRAM pool ... and I would want to see this exact configuration tested. 👀 192GB + OCuLink + dual 10GbE in 2.9L. Now show us the price... and somebody plug a 48GB Radeon into it. 🔥