Part 8. CPU SIMD Extension for X86 (Continued)
content by Xing Chen
AMD version AVX-512 appeared on Zen 4 series, and it is reported to be continued on Zen 5, “Additionally, the AVX-512 implementation with a full 512-bit data path ensures that floating-point and vector math operations [on Turin], essential for AI and HPC workloads, are handled with significantly enhanced efficiency.”, reported by StorageReview. I guess there will be more details leaked or published during this SC24 event.
Please refer to the attached first two pictures of Zen4 and Golden Cove microarchitectures, both are with the AVX-512 Extension.
Are they the same? Maybe? Not exactly?
Are they similar? Yes.
Well, thanks to the author(s) of these two images, because he/she/they has/have drawn the two images the same way, and that helps us to understand.
Where is the AVX-512 extension? Can you find it?
Well, maybe have a look of the third microarchitecture image, it is just a different way of drawing the Golden Cove microarchitecture, the AVX-512 is the block marked “VEC” for vector on the image. If you are really a hardware hacker, you can have a look of the zoom-in image attached and dig into more details by following this link: https://chipsandcheese.com/p/amds-zen-4-part-1-frontend-and-execution-engine
