AMD Ryzen AI Max+ Pro 395 Strix Halo Chip Breaks Cover With Radeon 8060S GPU

It isn't every day that a major chipmaker launches a whole new class of product. Intel arguably did this with its "Lunar Lake" processors, the first consumer x86 chips with on-package memory. AMD's now set to do it too, but instead of on-package memory, the company is launching an absolutely massive "APU."

AMD doesn't really use the "APU" nomenclature anymore, but greybeards like this author will recall the term describing CPUs with very powerful integrated GPUs for the time. Nothing fits that bill like the upcoming "Strix Halo," likely to be known officially as the Ryzen AI MAX family when it arrives early next year.


So what makes Strix Halo qualify as a new class of product? How about the fact that it includes a GPU with no less than 20 WGPs? For context, that makes the GPU approximately the same size as a Radeon RX 6750 XT, but it's likely to be faster than that old RDNA 2 part because it will be based on the RDNA 3.5 architecture, a refined and optimized version of the RDNA 3 design used in the Radeon RX 7000 series.

It's also unusual among APUs for having 16 full-fat Zen 5 CPU cores. Historically, AMD APUs have been based on monolithic mobile silicon with a limited number of CPU cores. Strix Halo, on the other hand, uses a pair of the very same CCDs that AMD uses in its desktop Ryzen and EPYC processors. In essence, it's kind of like a Ryzen 9 9950X stapled to the side of a powerful Radeon GPU.


Today we have our first leak of a presumably-final Strix Halo part (judging by it having its full name) on an AMD MAPLE test platform. The "AMD RYZEN AI MAX+ PRO 395 w/Radeon 8060S" showed up in the Geekbench 6 Compute benchmark database early this morning, and it's a whopper of a chip. Just as we said above, it has 16 Zen 5 CPU cores running at up to 5.1GHz, and a massive Radeon GPU delivering a benchmark score of 67,004.

That's not actually a great result, but beyond the fact that this is pre-release hardware running early software on a test platform, Geekbench reports that the memory transfer rate is a measly 1,994MT/s. Earlier reports indicated that Strix Halo was intended for use with quad-channel LPDDR5X memory running at 8,533MT/s; the Geekbench result confirms the quad-channel operation but the speed is catastrophically low, which we can probably attribute to it being a test platform.

So saying, we expect that performance out of a final-release Strix Halo chip will be much better on this and other benchmarks, but it's really exciting to see the earlier specification leaks confirmed. This is looking like an APU that can really go head-to-head against Apple's Mx silicon and we're dying to see what applications AMD and its partners have cooked up for such a caked-up CPU-and-GPU combo.