AI Discussions

[NotZeke]

I'm not sure if anyone knows the details on this GPU, as on the spec sheet there is no mention of AI, but on the press release and the RX 7000 series page it does mention AI accelerators on all RDNA3 GPUs. So I was wondering how much AI Tops it has and more specs on the AI accelerators in it. Thank you!

Also on a side note, what is the difference between a NPU's AI Tops and AI Tops from a GPU. Like I see many GPUS with high AI Tops, way higher than those NPUs, why have an NPU when you can use a GPU with AI accelerators?

[Uday_Das]

NPU provides acceleration with lower power, and sometimes for latency critical model better latency. For short running Model GPUs are good, but for longer running models NPU provides better performance per watt. They can both be used in different usecases. 

Thanks 

[Mary319Horne]

Hello!

You're right to notice the AI capabilities in AMD's RDNA 3 (RX 7000 series) GPUs like the RX 7900 XTX (around 123 INT8 TOPS) and RX 7900 XT (around 103 INT8 TOPS), which leverage AI Matrix Cores for accelerated machine learning tasks. While GPUs offer high raw AI performance, NPUs (Neural Processing Units) are specifically designed for power-efficient, low-latency AI inference at the edge, ideal for everyday AI features in laptops. GPUs excel at heavy AI training and large-scale inference, whereas NPUs handle lighter, always-on AI tasks without the high power draw, ultimately complementing each other in modern computing.