facebook rss twitter

Microsoft Silicon may be coming to datacentres and Surface

by Mark Tyson on 21 December 2020, 10:11

Tags: Microsoft (NASDAQ:MSFT), ARM, Surface

Quick Link: HEXUS.net/qaepxl

Add to My Vault: x

Please log in to view Printer Friendly Layout

Just ahead of the weekend Bloomberg published a report, citing unnamed insiders, saying Microsoft is working on its own Arm-based processors for servers and its Surface line of computers. It is important to understand the scope of the touted server processors – they are being designed for Microsoft to run its own cloud services on – not for third party use. On the 'Microsoft Silicon' Surface rumours, these plans seem to be less advanced at this stage, so devices like the Surface X look likely to stick with Qualcomm for at least another generation.

Microsoft's cloud computing rivals like Amazon, with AWS, have already established their own processor designs in their servers, so Microsoft's initiative isn't a huge surprise. Over recent years and months Microsoft has been strengthening its computer processor engineering talent, picking off personnel from the likes of AMD, Intel, Nvidia and Qualcomm. Of particular note is that it welcomed several of Qualcomm's chip designers to the fold after the smartphone chip designing giant abandoned server processor efforts.

Within Microsoft the chip design unit is said to report to Jason Zander, head of the Azure cloud business, rather than Panos Panay, who oversees Surface products. This points to the close partnership and primary purpose of any new Microsoft Silicon, but upcoming generations of Surface computers could also leverage Microsoft's own Arm-based SoCs, say the sources.

Microsoft partnered with Quaclomm for the SQ1 chip in the Surface Pro X

Since the initial publication of Bloomberg's report on Friday it has been updated with a statement from Microsoft that confirmed investments and developments in silicon design. "Because silicon is a foundational building block for technology, we’re continuing to invest in our own capabilities in areas like design, manufacturing and tools, while also fostering and strengthening partnerships with a wide range of chip providers," said Microsoft spokesman Frank Shaw.

Intel shares took another knock on Friday, down by over 6 per cent, with Bloomberg's story likely a contributing factor in the slide.



HEXUS Forums :: 8 Comments

Login with Forum Account

Don't have an account? Register today!
making your own cpu does not necessarily mean money will be saved, off-the shelf CPUs have a massive advantage: COMPETITION. Simply put the fastest is picked.
lumireleon
making your own cpu does not necessarily mean money will be saved, off-the shelf CPUs have a massive advantage: COMPETITION. Simply put the fastest is picked.

It sounds like Microsoft are making their own SoC, not designing the CPU cores themselves. They can still buy off the shelf core designs (probably N2 or even V1 cores), and pair them up with things like off the shelf 25GbE network cores to make something that properly fits their needs.

But for me the interesting thing here is whether the chips end up running Windows, or MS just put the usual datacentre standard of Linux on them.
DanceswithUnix
It sounds like Microsoft are making their own SoC, not designing the CPU cores themselves. They can still buy off the shelf core designs (probably N2 or even V1 cores), and pair them up with things like off the shelf 25GbE network cores to make something that properly fits their needs.

But for me the interesting thing here is whether the chips end up running Windows, or MS just put the usual datacentre standard of Linux on them.

Now that ARM are doing server orientated designs, most big datacentre providers will be considering this. Certainly Azure like AWS is big enough to afford to do so.
I'd expect Google to follow suite (think currently they only do those AI focuses TPU things).

As for MS and Linux, well they have allegedly become a bit more open recently (although the bootloader shenanigans on Win10 continue to give dual booters trouble), but on the other hand they have the Windows source code so could make kernels etc. which suite their workload.

Wonder whether the rumours about Apple doing some 8/16/32 core monster chips is mostly about datacentre too with maybe a use in a new consumer toaster (Mac Pro)? Otherwise, the market for their ‘pro’ stuff (which is generally anything but pro with limited expansion etc.) just seems far too small.
kompukare
Wonder whether the rumours about Apple doing some 8/16/32 core monster chips is mostly about datacentre too with maybe a use in a new consumer toaster (Mac Pro)? Otherwise, the market for their ‘pro’ stuff (which is generally anything but pro with limited expansion etc.) just seems far too small.

The thing is Apple will want something for devs to code on. As long as people are cross compiling from amd64 chips, ARM will be a second class environment. To suck the devs in, you need something at least vaguely on par with a high end AM4 chip.

This does all make me wonder how quickly AMD could dust off their ARM ideas if needed. A Zen 3 back end pipeline with ARM instruction decoders at the front could be a very rapid bit of silicon.
DanceswithUnix
The thing is Apple will want something for devs to code on. As long as people are cross compiling from amd64 chips, ARM will be a second class environment. To suck the devs in, you need something at least vaguely on par with a high end AM4 chip.

This does all make me wonder how quickly AMD could dust off their ARM ideas if needed. A Zen 3 back end pipeline with ARM instruction decoders at the front could be a very rapid bit of silicon.
Thing is who is actually left doing their own ARM designs currently?
For consumer / phone loads Qualcomm gave up, as did Samsung. HiSilicon might have been tempted that got Trumped.
So that leaves what? Only Apple really.
Server and that Fujitsu supercomputer is a different matter.
Still, neither x86 vendor can afford to ignore all this. Although for semi-custom consoles, I'd imagine AMD will have to be able to provide some AI acceleration for the next gen consoles I would think. Assuming someone can make some kind of AI framework for games - enemies with decent AI would be nice but the likes of Nintendo would probably come up with some custom controllers with enough image recognition circuitry.