From PC World: With the big AI providers cracking down on usage limits for their most powerful agentic features, it’s tempting to kiss our ChatGPT, Claude, and Gemini subscriptions goodbye and go all-in with local AI–and it’s certainly doable, if you’re willing to pay for it.
AMD is the latest to step up to the plate with a local AI solution, a mini PC dubbed “Ryzen AI Halo.” At first blush it looks mighty appealing: a Mac mini-sized form factor, a powerful AMD Ryzen AI Max+ 395 processor with 16 Zen 5 CPU cores and 32 threads (upgradable to the bleeding-edge Ryzen AI Max+ 400 series), a whopping 40 Radeon 3.5 GPU compute units, and–most important of all–128 GB of unified LPDDR5x memory.
That final specification is the key one. Memory is everything when it comes to AI inference (perhaps you noticed the ongoing RAM shortage), and without enough RAM, your system will struggle with large local LLMs like OpenAI’s 120-billion parameter GPT OSS, not to mention RAM-hungry video-generation models like LTX 2.3.
With its support for unified memory–that is, a shared high-speed pool of system RAM and VRAM–the 128GB AMD Ryzen AI Halo (which was initially teased back in January during CES) boasts a crucial advantage over discrete GPUs, which are limited by their separate VRAM stashes, generally in the 16GB, 32GB, or (if you can afford it) 48GB range.
View: Full Article
