AMD Optimizes Zen 4 CPU and Achieves +41% Performance with AVX-512

Currently, the most powerful consumer processors we have for PCs would be the Intel Core i9-13900KS and the AMD Ryzen 9 7950X. While both CPUs are often targeted for purchase by gamers, they can also achieve great performance in other applications and programs. It is now reported that the latest AMD processors have received a compiler update, which is now supports Zen 4 and AVX-512which allows the Ryzen 9 7950X achieve a significant performance improvement.

The AVX-512 vector instruction set has divided quite a few users on whether they are really useful or wasteful. And it is that, the processors that make use of these, see their consumption and temperatures skyrocket. This to the conventional user who only uses the CPU to play and office tasks can become a headache, since used by various programs. In fact, AVX-512 used for computation, conversion and processing of images and videos either cryptography. We have always related to Intel with this set of instructions, but AMD with its Ryzen 7000 already had AVX-512, yes, through two 256-bit FMA.

AMD updates its compilers and increases the performance of AVX-512

When the Ryzen 7000s were released, support for AVX-512 was not showing the results that were actually expected. Although they were significantly better than the Ryzen 5000, AMD promised figures that were not achieved in practice. But at the end of 2022, the company launched the version 4.0 of their AOCC and AOCL compilerswhere they finally have support for the Zen 4 architecture. So, at Puget Systems they have made a review to see the Ryzen 7000 performance with AVX-512 in this new version.

Using a PC made up of a 16-core Ryzen 7950X, 64 GB of DDR5 4800 MHz RAM, the X670E AORUS MASTER motherboard and the Ubuntu 22.04 operating system, let’s proceed to see the results. Starting with HPL Linpack, the Ryzen 7950X with Zen 4 and AVX-512 optimizations got a score of 1011 GFLOPs/s, just over half of a Threadripper Pro 5995WX with 64 cores. Compared to the previous results, we see a clear improvement compared to 916 points and 847 points with the old versions. In fact, there is a jump 45% compared to the Ryzen 9 5950Xwhich gets 695 GFLOPs/s in this test.

Ryzen 9 7950X with Zen 4 optimizations nearly doubles Ryzen 5950X

We now turn to the HPCG benchmark, which is limited to memory and used in engineering simulations. Here we see the big difference between Threadripper and desktop AMD Ryzen. And it is that, the AMD Threadripper Pro 5995WX with 8 memory channels manages to achieve 19 GFLOP/sMeanwhile he Ryzen 7950X with Zen 4 optimizations of AVX-512 is limited its performance with 10.53 GFLOPs/s. We would be comparing 8 channels vs. 2 memory channels and even so, less than half is lost. Furthermore, if we compare it with the 7.46 GLOP/s of Ryzen 7950X without the optimizationswe see an incredible 41% improvement.

In fact, almost double the 5.76 GLOP/s of Ryzen 9 5950X what you achieve on this test. Finally, the test of HPL-MxPa benchmark for solving linear equations, which reflects the performance in ML/AI inference and training models. Here we see that the Ryzen 7950X with AMD’s AVX-512 optimizations achieves a performance of 3.975 GFLOPs/s. Although there are no comparisons, Puget Systems acknowledges that this is a very promising result, since they are almost 4 TFLOP/s of mixed precision.