In multiple interview sessions on Thursday, AMD director of technical marketing Robert Hallock answered many questions from journalists and tech enthusiasts about the company's 5nm Zen 4 'Raphael' Ryzen 7000 processors. One such interview, part of the 2.5 Geeks series from Hot Hardware included Hallock's confirmations of launch core counts, that there are new Threadrippers coming, and the heavy hint of Ryzen 7000's support for AVX-512 instructions.
AMD's communications regarding its 5nm Zen 4 'Raphael' Ryzen 7000 processors and AM5 socket motherboards didn't get off to the clearest of starts. However, we have now pretty much ironed out all the wrinkles from the announcement. Two of the biggest areas of uncertainty after the Computex keynote were concerning Ryzen 7000 power specs, and CPU performance vs IPC.
Hallock touched upon the 15% single-threaded performance uplift claims made at Computex keynote. He told the site that the provided figure was chosen so that AMD wouldn't have any chance to disappoint, with "oftentimes a lot more [performance]," than that figure.
Hallock confirmed that Zen 4-powered AM5 processors will max out at 16 physical cores at launch. Of course, current-generation Ryzen chips already reach these core count heights (e.g., Ryzen 9 5950X). However, Hallock was keen to point out that "we're getting 40%+ more performance out of that core count." Moreover, he teased that the future holds even greater promise.
There is also a short but sweet section of the video regarding Threadripper processors. When asked about these HEDT processors, which were conspicuous by their absence from the Computex presentation, Hallock said that there are "more coming." We recently reported on a shortage of Threadripper processors in the channel – hopefully, this is a good sign that AMD is preparing for a refresh.
On the topic of AVX-512 support, Hallock refused to be drawn into an official answer. The interviewers did their best to get the AMD technical marketing lead to name this particular instruction set instead of referring to the official "Expanded instructions for AI acceleration." Their efforts, sadly, fell a bit flat.
However, in an interview with TechPowerUp published earlier today, we have a much clearer statement about the relationship between AMD's expanded instructions supported by Ryzen 7000 and AVX-512. Asked specifically about this relationship, Hallock told TPU that the new Ryzens specifically support "AVX 512 VNNI for neural networking and AVX 512 BFloat16 for inferencing." Furthermore, it was explained that this implementation delivered some "pretty nice speedups" in tasks like video upscaling, which is quite a common endeavor for enthusiasts.