[ad_1]
A few weeks ago, Apple released the ARM-based M1. The new chip has set performance-per-watt efficiency records and has shown that it is capable of drilling well above its weight class. Now, there are credible rumors that Microsoft plans to build its own ARM cores for servers and PCs. As always, take your rumors off with a little sodium chloride.
The M1’s biggest threat to the x86 market has always been the products it could inspire other companies to build. For decades, we have operated under the assumption that x86 CPUs represented the fastest CPUs you could build. That assumption was based on some specific facts:
- Almost all non-x86 vendors that tried to compete with Intel on desktops, mobile devices, or servers died.
- AMD, Intel’s only long-term historical competitor, has typically only led the market for short periods of time. Intel has always regained leadership in the long term. Intel has been the company that has set the pace for improvements in consumer CPU performance for most of the past two decades.
- Intel itself had tried to break free of x86 and establish a new family of CPUs on multiple occasions. It failed all the time. One of the reasons it failed is because engineers kept finding more ways to improve the performance of x86 CPUs.
With no non-x86 competitors to compare against, and Intel has tried repeatedly and failed to break out of the standard in the first place, there was no perceived alternative or openness in the market. AMD could challenge Intel specifically on x86, but AMD has never had the market leverage to drive adoption of an entirely new CPU architecture. AMD is specifically valued because it offers a perfectly compatible alternative to Intel.
Then Apple released the M1 and proved, for the first time in decades, that it was possible to outperform x86 CPUs on their own ground. Anandtech has a review of the Ampere Altra Q80-33 server platform at the moment. It packs 160 cores in total, and while it doesn’t clearly outperform the Epyc 7742 in all benchmarks, it does demonstrate a lower overall TCO, excellent performance in many areas, and a clear path for future ARM server development. Nuvia should make more disclosures about its own ARM CPU in 2021, and Apple is expected to introduce a 16-core M-class core as well.
We have raised the idea that Microsoft could build its own SoC to compete against Apple with Surface devices, especially if AMD and Intel cannot provide equivalent competitive hardware. Bloomberg sources believe the company will push to build a server chip first before a consumer-oriented solution, but the idea that a major company like Microsoft is looking to build its own core to reduce reliance on manufacturers like Intel is not surprising. .
I don’t want to imply that the M1 is the only factor at play here because it is not at all. Intel’s data center sales have exploded in recent years, but its CPU architectures have largely stalled. AMD has decisively outperformed the company in terms of cores per socket, and that is not expected to change dramatically anytime soon.
As always, I take this information with a grain of salt, but I can’t say that I was surprised to see it. Intel’s repeated missteps since 2014 have diminished confidence in the company’s ability to deliver the fastest and most powerful CPUs on the market. ARM chips have proven to be true contenders. AMD has moved the x86 market forward much faster than Intel, but AMD is still building its presence in the server market.
It takes time, usually 3-5 years, to build a new core from scratch, so I wouldn’t be looking for a new chip from Microsoft in 2021. As with Apple, AMD and Intel have some time to further improve their own products. And of course, just because Microsoft is designing its own CPU core does not automatically mean that the core will be competitive.
But if this rumor is true, it’s one more thing for both companies to worry about. We will see more such announcements in the coming years.
Now read:
[ad_2]
