Local inference just got real. AMD dropped a mini workstation under four grand.
I've been running models through cloud APIs for about two years now and the costs add up fast when you're doing anything beyond basic prompts. Like genuinely painful once you scale past hobby projects. Was sitting in my home office last Tuesday staring at another monthly bill and just thinking there has to be a better way. So seeing a compact box that can handle local model runs at roughly the same price point as a decent gaming rig, that changes the math completely.
The NVIDIA alternative sits around forty seven hundred. Not a massive gap on paper but when you factor in that the AMD unit runs both Windows and Linux natively, the flexibility alone makes it more interesting for most dev workflows I've seen. And its like Mac Mini sized which is kind of absurd for what it does.
Cloud bills might actually have competition now.
[link] [comments]