Any comments from Intel on the M1 Ultra?

[thedocbwarren]

Yes, they will come out with something using more cores and more power to edge out the M1 Ultra, which will mean absolutely nothing to anyone.

Using liquid nitrogen to cool it no doubt and run for less than 500ms.

 

[Unregistered 4U]

Again I just don't get the point of contention here. Of course a new CPU will require scheduler changes with an OS to fully take advantage of said CPU.

There are several sites that are calling it out as something that hasn’t occurred for prior Intel releases, and they’re right. Wouldn’t be surprised if this on the roadmap was part of the reason for the Apple bowing out as well. “Hey, Apple, yeah, we haven’t really been providing the kind of processors that you need for years now BUT, it would really mean a lot to us if you could do us a solid and implement these changes in your scheduler <slides stacks of papers across the table> because, it’s just TOOO hard for us to do schedule alone what with the hyper threading not being on all of the cores and such.” The more I look into it, the more I agree with the SemiAccurate’s assessment that indicates Intel’s implementation is needless complexity.

The argument that this is a potential problem for Intel because Microsoft will stop bothering to optimize their OS for X86 because they'll focus on ARM Windows going forward - an OS variant that currently has fractions of a single % uptake, is bizarre. Sure, perhaps in a hypothetical future decades from now? I mean what's the point of even speculating that far out.

I’d guess it’s similar to how folks were speculating years ahead of Apple’s eventual switch to their own processors. It’s a rumors site! And, Intel getting shafted by first Apple and THEN Microsoft? That would be a juicy rumor indeed!

 

[Fragment Shader]

I’d guess it’s similar to how folks were speculating years ahead of Apple’s eventual switch to their own processors. It’s a rumors site! And, Intel getting shafted by first Apple and THEN Microsoft? That would be a juicy rumor indeed!

Hardly the same thing. People were speculating Apple would switch Macs to their own CPU's because they had sold hundreds of millions of computing devices with their own CPU's by that point. It was "when" and not "if" for years before the M1 debuted. Apple just didn't flip a switch, they had years and years of engineering wins under their belt before the Mac finally jumped on board.

For Microsoft, you've got...the Surface Go + Duo, with generally negative reviews and a basically non-existent market share in the categories they're actually in. Sure this is a 'rumours' site, but not all hypotheses deserve the same level of discussion if they're not remotely tethered to any real-world conditions that currently exist. You're speculating on something that could potentially occur by the time Alder Lake has been discontinued for a decade.

Like ARM requires a hell of a lot more work to get 'support' for Microsoft than just the scheduler changes for Alder Lake, Microsoft has to devote significantly engineering resources to build a decent x86 emulator for ARM, and in fact is helping to design their own ARM CPU's in part for this purpose (Apple's M1's are so adept at running X86 code in part due to their specific hardware that assists in this).

When the theoretical day comes that an X86 emulator in Windows ARM isn't necessary because native Windows ARM apps are plentiful, Intel's architecture will likely differ far more significantly than Alder Lake does to previous X86 implementations today.

 

[killawat]

Hardly the same thing. People were speculating Apple would switch Macs to their own CPU's because they had sold hundreds of millions of computing devices with their own CPU's by that point. It was "when" and not "if" for years before the M1 debuted. Apple just didn't flip a switch, they had years and years of engineering wins under their belt before the Mac finally jumped on board.

Don't forget the iPhone. Johny brought us Apple silicon since the A4. The Mac team had probably been begging for his expertise, saw what they were able to do with the T-Series chips, and finally made the leap to the M series chips. So many chips, but no dip!

 

[Krevnik]

And macOS 11 shipped with a scheduler for M1's big.LITTLE design... what's the difference?

Not a huge one, but I think the key difference in my mind is that Apple’s scheduler for iOS/macOS is surprisingly braindead. Like classic CPU designs, tasks start with the “primary” core and work towards the bottom of the core list as more resources are needed. E cores for Apple are always at the bottom of the list, and tasks that are “background” priority (not necessarily an app in the background, the priority is different than active/inactive state) are limited to only run on these cores. That’s it. That’s the trick. Apple’s scheduler looks a lot like CPU schedulers today.

SMT on Intel isn’t too bad either. The logical cores that represent the second thread on each core live at the bottom of the core list, so the cores are loaded first before the second thread of any core receives work. So the same sort of scheduler that works for SMT represents most of what Apple’s custom scheduler does.

However, with Alder Lake, you have Thread Director, and then logic in Windows which is looking at things like process being active/inactive, on top of the SMT logic, and having to know where the P/E cores start in the core list. All because Intel’s goal here is to move threads between the P and E cores, which Apple doesn’t really do.

Generally for something like a scheduler, the simpler the better as it keeps it reliable.

 

[waloshin]

Intel compete even with upcoming Ryzen 7000... ?

Even modern Intel CPUs are losing up to 36% performance with the new Spectre BHI patch

A new CPU exploit based on the infamous Spectre v2 has been discovered. However, when the security patch is applied it can affect performance by up to 36%, that's according to a recent study.

www.neowin.net