AMD Zen 5 Architecture will Introduce a New Instruction Sets for Enhanced AI Performance

Phoronix has recently shared that the upcoming release of the GNU Compiler Collection (GCC) is getting ready to embrace AMD's next-gen Zen 5 architecture. It's exciting to hear that Zen 5 will bring with it at least five new instruction sets, and one that's catching eyes is the AVX-VNNI. Interestingly, four of these instruction sets are already familiar faces in the Intel CPU world, and there's buzz about another one joining the ranks in Intel's upcoming Granite Rapids server CPUs. Historically, AMD has been a bit slower than Intel in rolling out new instruction sets, especially those related to AVX.

Digging into the details, AVX-VNNI was a game-changer first introduced by Intel with its Alder Lake CPU back in 2021. Standing for Vector Neural Network Instructions, it's a big deal in our AI-centric world today. The difference between AVX-VNNI and its big brother, AVX512-VNNI. Although they share some common ground in functionality, AVX512-VNNI pulls ahead with its 512-bit vectors and EVEX extensions, offering enhanced performance.

With Zen 4 already supporting AVX512-VNNI, the addition of AVX-VNNI to Zen 5 might not seem like a headline at first glance. However, it's an important step, especially for consumer-level processors, reflecting a strategy Intel has also employed in the past.


Printed from: