AMD is one of the main companies participating in Computex 2023, the popular technology trade show held annually in Taiwan. In light of the growing popularity of artificial intelligence services, the hardware giant reinforced its advances in the area with a demonstration of the power of its Ryzen AI-powered processors.
During the event, the company talked about the process of preparing the entire software ecosystem for the local processing of tasks based on artificial intelligence on PC. For this, AMD highlights its new AI processing engine integrated into Ryzen. Known as “XDNA AI”, this technology was demonstrated in practice at the Taiwanese fair.
Courtesy of Tom’s Hardwarethe image above shows XDNA AI running on an ASUS ROG Zephyrus G14 notebook equipped with a Ryzen 9 7940HS processor and integrated graphics based on RDNA 3. In the test in question, the artificial intelligence engine performs facial recognition work on photographs .
It can be seen that the bar latency measurement remained remarkably low during testing. The inference workload processed a series of images quickly, proving that local processing is feasible and convenient.
The brand’s booth tried to compare Ryzen AI processing with the Azure ONNX EP service, which runs on Microsoft servers, but the team found connection problems and the test had to be put aside. The goal was to reinforce the low-latency advantage of local processing against cloud operation.
In any case, while we don’t have context on how the numbers compare to other types of solutions, it’s clear that Ryzen AI is increasingly robust and ready for a possible new era of computing.
XDNA AI handles up to four concurrent AI workflows currently, but that number could easily increase, according to AMD. A highlight of this technology is its low power consumption. The manufacturer claims that its cores dedicated to AI are up to 50% more efficient than the Neural Engine of the Apple M2.
AMD is making the job easier to add software support to XDNA AI. David McAfee, general manager of the company’s client business, cites the Vitis AI Execution Provider, which serves as a “translator” that allows developers to run models without having to change the base model.
This simplifies application integration and compatibility. AMD’s implementation will work with the same applications that Intel uses with its VPU integrated into the “Meteor Lake” family of processors, such as the Adobe suite.
McAfee adds that while AMD is committed to the AI engine, it is not yet expected to make its way into all of its products, such as desktop chips, which are currently the target of other brand strategies such as chiplets.
