So it’s been confirmed. Alder lake will have “little” and “big” cores simultaneously on the same chip. Personally? I don’t like it. If I’m building a PC, I want a fecking DESKTOP CPU. Not half of one, with the other half essentially being a mobile CPU. Apparently the lower power cores are Atom cores…?
This is literally just ADDING to the confusion. Their nomenclature has become confusing. Their process nodes have become confusing. Everything about this company has become so confusing in just the past 3-4 short years it’s making my head hurt.
But more to the point; I don’t like the idea of buying essentially half of what’s supposed to be a desktop CPU. Maybe it will surprise us? I’m already seeing the floodgates open for fellatio-contest news articles regarding the new CPU. “Biggest performance increase in 14 years”. Supposedly…
Knowing the tech world. I’m going to call that statement as another form of hype generation/indoctrination. Again i’m not saying this is set in stone; we could be pleasantly surprised? But… Just on paper: Lower powered cores, in unison with half of the total cores being true desktop cores. This sounds like it could be VERY beneficial in SPECIFIC SITUATIONS. I believe there is quite a bit of specificity involved in how this could deliver in terms of performance. But overall? I don’t see this panning out.
And on a side note. If any of you remember my plans when I built my current PC 4 years ago… WOW. I am REALLY beating myself up for what I did. Me: Oh… Intel will come back with their answer to Ryzen and it will be their next Nehalem!
This may be relevant in mobile. Where intel have repeatedly failed.
For server/desktop/laptop? Irrelevant/useless. Especially given the hybrid cores are using different instruction sets. It’s not just that they run slower. They don’t even have the same feature set of the big cores. Which means they can’t run the same code.
It’s going to be a software scheduling nightmare (e.g., code is spun up on low power core and hits an instruction the little core doesn’t support**), and microsoft can’t even get that right on relatively traditional X64 cores (see: thread ripper).
Do intel really think microsoft are going to completely re-architect the Windows scheduler to handle this shit? And if so, do they really expect that to happen in any sub 3-5 year time-frame?
again
edit:
** actually I think maybe intel is disabling their “magic instructions” entirely (?) (to paraphrase Linus Torvalds) on Alder lake because the little cores don’t have them.
On the one hand they’re pushing AVX512 on Ice Lake and on the other disabling support on Alder lake. Intel have no idea what they’re doing, no clear path forward and are flailing wildly in panic.
Ive been meaning to watch this, thanks for the reminder
Intel Xe is another misadventure they will throw colossal amounts of money at, only to lose to Nvidia.
I am certainly not going to get one of the alder lake chips. However, I would like to see a really good desktop or workstation chip with some hybrid cores. But not the way intel is doing it. I would love a chip with one or two little cores that can do idle and background stuff really well, at very low power. And about 8 or more big cores for when I am doing more than just scrolling memes.
Honestly though, I think my solution is to just get an 8gb ras pi as the main desktop and remote into my threadripper when I need power.
From the looks of the Lakefield one that is out now, they will be terrible. I also think the approach of letting the OS manage which core does what is not a great idea if they are targeting windows, sure Microsoft can do it but it is a question of will, when and how good or if it will even be a standard spin of windows 10 or one of the many forgotten experiments.
IF AMD manage to pull it off how they are describing and properly with the current Ryzen stuff, the combination of the low power laptop cores, high power desktop cores and some Navi for good measure they could make a really nice little system.
I liked the idea of the V1000s but they were a little bit industrial, I have a “Gaming” NUC and enjoy that. If AMD made a “NUC” style tony brick with some half decent graphics and hybrid cores I would be all for it.
Correct. But it will contain the hellfire furnace that has been their problem for half a decade. Intel can no longer make a low power, high clock, high IPC, instruction heavy, AVX laden CPU core that ticks all the boxes. They can build cores that do some of those things really well. So mix and match to find the balance.
Intel’s oldest running prayer: If we build it, compilers will improve.
That aside Windows has worked on a ton of obscure platforms and the scheduler is practically modular at this point. By no means perfect, but in 2017 and 2018 they unfucked a lot of it purely because they were forced to by Windows on ARM projects and more importantly issues with Ryzen and Threadripper. I’ll say from just messing around at a surface level I did see the scheduler in Win10 1809+ significantly improve Bulldozer scaling over all previous versions. Unrelated to big.LITTLE but still an impressive side note considering how wide-scale NUMA enabled Bulldozer deployments were considered ‘inconsolable’ and improvement attempts abandoned for 6-7 years.
Not really, it will still be a hellfire furnace if actually running code that needs to run fast, where the dead-weight little cores are worse than useless.
That would be complete contrary to the design goals, and almost impossible to actually fuck up that bad. Like you’re expecting those four 1.5W Atom cores to contribute anything in terms of thermals to the 12.5W core while it’s at peak turbo during tau? No they’ll be sipping along at half utilization keeping background tasks and I/O out of the way while the primary core(s) handle the heavy stuff. Not to mention idle and dark silicon do assist in thermal distribution and dissipation.
I mean I get it; Intel man bad. Can’t have any competent engineers among the 110,000 people in the company.
As opposed to just including another big core that can actually run the instruction sets they are trying to push to speed up common tasks at a lower clock speed within similar power envelope.
Like I said, desktop - they’re worse than useless. Mobile - that ship has sailed and aint coming back.
intel should learn from their own history:
people run x86 for compatibility reasons on legacy code.
people are not going to massively change schedulers/application optimisation/etc. for some oddball intel architecture when they can just buy something that already runs their existing code faster. It’s totally not clear how intel expect all the APPS out there to deal with the thread optimisation within the app for this.
They tried to get away from “standard” x86 with Itanium. it tanked. this idea will tank as well, because if people are going to go re-optimising/rewriting code in order to make it run better in a power sensitive environment, they’re probably just going to port it to ARM instead - where the majority of the market already lives, and where there’s the ability to license and more easily customise the processor to your specific requirements. not what intel are pushing.
You do realize that those little cores are x86-64… Right? They’re Gracemont cores, an evolution of Tremont with all the widely adopted SIMD extensions still in place. And you also realize MS has been rebuilding their scheduler for HSA since 2017, so the support for this approach already exists on the software side regardless.
They’re x64 sure, but they don’t include AVX and other instructions intel are trying to push elsewhere for performance reasons. And thus those instructions are also disabled on the big cores according to the anandtech article I read on it.
Which means that all the AI deep learning boost, etc. that intel are pushing at the moment with ice lake, etc. will not work. At all. Even on the big cores.
AVX is underutilized at best anyway. As stated previously these are not chips designed to check every box in the feature list. AVX is a massive resource hog and is the root of most of Intel’s current thermal issues. Most AVX accelerated workflows will still work with SSE as a fallback, and even shit like MacOS which is AVX native from EFI still has an SSE2 runtime that it can default to.
As an aside; Intel is where they are right now because they took zero risks. They need to step up and start trying new shit or they will continue to fall.
AVX clock speed goes down by only 200-300 mhz on ice lake. So its not as bad as it was, moving forward - and intel have committed to that path - so commit! You’re not going to get improvements from it if you cripple it on half your product line!
intel is where they are right now BECAUSE they risked too much on 10nm. They went for 2.4x scaling, which was insane after the problems they had getting 14nm out the door originally.
The simple fact is that Intel need a manufacturable node improvement beyond 14nm even if it is relatively mediocre. They’re way behind today and by the time they get 10nm out at scale TSMC will be on their 5nm node and already working on 3nm. Even if we give intel the benefit of the doubt on intel 10 vs. TSMC 7 (which imho is basically lies - based on my 10nm MBA vs. 10nm older iPad Pro, I can tell you now the iPad Pro CPU from 2018 on TSMC 10nm slaughters ice lake in real world performance performance in less power with no fan running the same apps (e.g., swift playgrounds which runs on both)).
Wouldn’t surprise me if intel big.little is more about being able to get 10nm to actually yield (there’s a reason desktop isn’t on it yet). They can fuse off broken AVX units on the big dies and the little dies will be easier to get to yield.
Attempting to make lemonade… intel can have the best CPU design engineers in the world (arguably they don’t any more, but let’s play devils advocate and assume for the moment they do); currently they’re getting fucked by the muppets running their production. You simply can’t make up the ground they’ve lost with CPU design at this point IMHO.
This is a large part of how intel put the other guys out of business - foundry process leadership. Now the shoe is on the other foot.
