It is so nice to see a return of competition in the CPU space. For too long it had been a one-horse race, with Intel on its own and AMD willing to settle for good enough. Revitalized with the Zen architecture, AMD is taking it to Intel once again, and you are the winner.
Both sides are proclaiming massive performance records, although in both cases they come with an asterisk next to them.
In Intel’s case, it announced 95 new performance world records for its Intel Xeon Scalable processors using the most up-to-date benchmarks on hardware for major OEMs, including Dell, HPE, ASUS, and Super Micro, running SPECInt and SPECFP benchmarks as well as SAP HANA, ranging from single-socket systems up to eight-socket systems.
Even more impressive is that all systems tested included mitigations for the Spectre and Meltdown vulnerabilities. To mitigate those flaws in the CPU, some functions have to be disabled either in software or at the firmware level, and that can mean performance hits, sometimes significant hits.
It shows Intel has a per-core advantage because its top Xeons are 28-core, whereas the AMD Epyc is 32-cores. And one benchmark shows Intel’s AVX512 extensions clobber Epyc, which enables floating point instructions and impacts compute, storage, and networking functions. So, it’s a big deal.
The down side? They might be hard to get. Last week CFO and interim CEO Robert Swan published a letter saying that the company was suffering from a shortage of chips due to increasing demand, but Swan assured customers the company would be able to meet demand. The good news for data center operators is that the problem seems to be more on the PC side than server side.
“The surprising return to PC [market] growth has put pressure on our factory network,” said Swan in a statement. “We’re prioritizing the production of Intel Xeon and Intel Core processors so that collectively we can serve the high-performance segments of the market. That said, supply is undoubtedly tight, particularly at the entry-level of the PC market.”
Then there’s AMD, which has to be feeling its oats right now. AMD and Xilinx announced they set an AI inference processing record of 30,000 images per second. It was done on a system using a combination of AMD’s Epyc CPUs and the new Xilinx Alveo FPGA accelerator card.
The announcement took place at the Xilinx Developer Forum in San Jose, California, and had an accompanying blog post. According to the blog post, “AMD and Xilinx have shared a common vision around the evolution of computing to heterogeneous system architecture and have a long history of technical collaboration. Both companies have optimized drivers and tuned the performance for interoperability between AMD Epyc CPUs with Xilinx FPGAs.”
This is something of a forced marriage, after all. Xilinx was one of the two dominant players in the FPGA market. The other being Altera, which Intel bought in 2015. So, as the only other CPU and FPGA makers, the two had nowhere to go but to each other.
And FPGAs are monsters when it comes to repetitive tasks, so the caveat here is this wasn’t a solo AMD accomplishment; it had a partner. Still, even if this is something of a shotgun marriage or forced alliance, it won’t hurt AMD to have an accelerator partner.
Once again, all in the name of increasing competition.