Hacker News Reader

> The Pro 9000 WX-Series processors are also designed for local AI deployments, including fine-tuning models, inference, and AI application development. “When running a context-based prompting inference test using DeepSeek R1 32B, we are seeing a 49 percent better performance of Threadripper Pro 9000 over Intel,” claims AMD.

Are they using the CPU as a GPU here?

View on HN

Yup, that's a thing. Mostly along the lines of "CPU memory is cheap, GPU memory is expensive, if CPU performance is made good enough we can run really really big models for much less because of cheaper RAM"

I'm all for it. It enables those really big models with a relatively small number of active parameters.

Reply on news.ycombinator.com