AMD has expressed its intentions to expand Ryzen AI throughout its entire product lineup, prioritizing areas where it can add the most value. At CES 2023, AMD unveiled Ryzen AI as a dedicated AI engine for its CPUs. The Ryzen AI co-processor, integrated on-die, enhances AI capabilities. This feature allows the Ryzen AI Engine to handle up to 4 concurrent AI streams while multitasking, delivering a 35% increase in responsiveness compared to a single AI stream.
However, there are considerations when it comes to incorporating Ryzen AI into chips, primarily related to cost and overall value proposition. AMD’s DVP & GM of Client Channel Business, David McAfee, explained in an interview with PCWorld that Ryzen AI is just the beginning of AMD’s AI endeavors. CEO Dr. Lisa Su has also emphasized AI as the company’s top strategic priority.
The initial chips to utilize the AMD Ryzen AI engine are the Ryzen 7040 “Phoenix” mobility processors. Not all Phoenix chips feature the dedicated AI engines, as AMD has made this decision based on certain factors. These chips, based on Xilinx IP, optimize software and operating systems like Windows 11 by enhancing hardware efficiency rather than burdening the CPU and GPU.
AMD is already discussing plans to expand the Ryzen AI stack to other Ryzen products.
However, the cost remains a significant hurdle, and there must be a compelling reason to include Ryzen AI in budget chips and even desktop SKUs. David also suggests the potential addition of Ryzen AI to a Threadripper chip with high core counts, primarily for training purposes, without a guaranteed implementation.
Software plays a crucial role in overcoming the cost and value barriers associated with Ryzen AI. As software evolves and better utilizes AI, offering increased value, the demand for dedicated AI hardware blocks on chips becomes more compelling.
Recent rumors suggest that AMD is developing a future generation of EPYC CPUs with an integrated dedicated AI chiplet. While the design structure may differ from the Ryzen AI block due to the EPYC’s nature, it will likely leverage similar Xilinx IP.
Also Read:
