AMD intros Ryzen AI 300 chips with Zen 5, better GPU, and hugely improved NPU

AMD’s next-generation laptop processors are coming later this year, joining new Ryzen 9000 desktop processors and ushering in yet another revamp to the way AMD does laptop CPU model numbers.

But the big thing the company wants to push is the new chips’ performance in generative AI and machine-learning workloads—it’s putting “Ryzen AI” right in the name and emphasizing the presence of an improved neural processing unit (NPU) that meets and exceeds Microsoft’s performance requirements for Copilot+ PCs. The new Ryzen AI 300-series, codenamed Strix Point, succeeds the Ryzen 8040 chips from earlier this year, which were themselves a relatively mild refresh for the Ryzen 7040 processors less than a year before.

AMD promises performance of up to 50 trillion operations per second (TOPS) with its new third-generation NPU, a significant boost from the 10 to 16 TOPS offered by Ryzen 7000 and 8000 processors with NPUs. This would make it faster than the 45 TOPS offered by the Qualcomm Snapdragon X Elite and X Plus in the first wave of Copilot+ compatible PCs, and also Intel’s projected performance for its next-generation Core Ultra chips, codenamed Lunar Lake. All exceed Microsoft’s Copilot+ requirement of 40 TOPS, which enables some Windows 11 features that aren’t normally available on typical PCs. Copilot+ PCs can do more AI processing locally on device rather than relying on the cloud, potentially improving performance and giving users more privacy.

If you don’t particularly care about generative AI, locally executed or otherwise, the Ryzen AI 300 processors also come with an updated CPU based on the same Zen 5 architecture as the desktop chips, plus an “RDNA 3.5” integrated GPU to boost gaming performance for thin-and-light systems that can’t fit a dedicated graphics processor. The chips are being manufactured on a TSMC N4 process.

AMD is announcing two chips today, both in the high-end Ryzen 9 series. The Ryzen AI 9 HX 370 includes 12 CPU cores and 16 GPU cores, up from a maximum of eight CPU cores and 12 GPU cores for the Ryzen 8040 series. The Ryzen AI 9 365 steps down to 10 CPU cores and 12 GPU cores. Both have the same NPU onboard.

Though an increase in CPU core count suggests big improvements in multi-threaded performance, note that in both chips a majority of the CPU cores (8 in the 370, 6 in the 365) actually use the “Zen 5c” architecture, a variant of Zen 5 that supports the exact same instructions and features but is optimized for small size rather than high clock speeds. The result is essentially AMD’s version of one of Intel’s E-cores, though without the truly heterogeneous CPU architecture that has caused incompatibility problems with some apps and games.

This isn’t the first time we’ve seen a mix of big and small CPU cores from AMD, but it is the first time we’ve seen it at the high-end. Zen 4c cores only really showed up in lower-end, lower-power CPU designs in the Ryzen 3 and 5 and Ryzen Z1 families.

Perhaps tellingly, AMD offered no direct comparisons between the CPU performance of the Ryzen AI 300 chips and the Ryzen 8040 series, opting instead to compare to offerings from Intel, Qualcomm, and Apple. This certainly doesn’t mean performance has regressed generation over generation, but it is usually code for “this isn’t the kind of improvement we want to draw attention to.”

AMD also didn’t offer performance comparisons between the new Radeon 890M and 880M and the old Radeon 780M. The company said that the 890M was an average of 36 percent faster in a small selection of games compared to the Intel Arc integrated GPU in the Meteor Lake Core Ultra chips and 60 percent faster than the Snapdragon X Elite in the 3DMark Night Raid benchmark (this was part of a slide that was specifically highlighting the performance impact of translating x86 code on Arm chips, though for the time being it’s true that the vast majority of games running on Snapdragon PCs will have to deal with the overhead of code translation).

AMD says that the Ryzen AI chips are slated to appear in “100+ platforms from OEMs” starting in July 2024, a month or so after Microsoft and Qualcomm’s first wave of Snapdragon X-equipped Copilot+ PCs. Ryzen AI will also compete with Intel’s next-gen Lunar Lake chips, also due out sometime later this year.