Intel Alder Lake mobile CPU with 14 cores spotted

[Wirko]

Ahh yeah I've just read that article, seems the issue may be with AVX-512, TSX and FP16 - I guess that could be more workable (and they're far less applicable to desktop for now anyway). I'd made the assumption that AVX2 was also affected, but still wonder how much impact scheduling will have - I assume the AVX2 pipeline is much narrower on Gracemont vs the big cores?

Unknown instructions can be emulated - they generate an interrupt, which can be trapped, processed, and the interrupted thread can then continue. It's inefficient but can prevent a thread from crashing until it can be moved to another core.
Probably not possible for TSX, though.

[JayN]

I'm still wondering what Intel intends to do with the pcie5 on the Alder Lake chips. That's more of a puzzle than Gracemont.

[six_tymes]

"potential pitfalls of hybrid processors coming to Windows." why end with such a negative statement? what is the agenda?

[lumireleon]

Destined to fail EVENTUALLY? M1 has a hybrid system, Intel is trying catch, not bad but will take a decade or more to mature. Programmers who make apps hate closed-source tech considering its Intel/Apple, unlike the ARM space where engineers were willing because ARM makes +95% of the smartphone market and Android is a killer. Will someone take resources just to create an App that works brilliantly only on intel? that's dangerous because there is ARM, AMD and M1. What intel should 'simply' do is to work with AMD, crazy but having an array of different hybrid layouts and implementations (2+1, 1+5, 6+8 etc) will spell nightmare for the operating system vendors and app engineers even GPUs will be affected. Is it time to update the X86 architecture? AMD is also to blame for not working on AVX512 (or is it plain marketing).

[Iota]

Destined to fail EVENTUALLY? M1 has a hybrid system, Intel is trying catch, not bad but will take a decade or more to mature. Programmers who make apps hate closed-source tech considering its Intel/Apple, unlike the ARM space where engineers were willing because ARM makes +95% of the smartphone market and Android is a killer. Will someone take resources just to create an App that works brilliantly only on intel? that's dangerous because there is ARM, AMD and M1. What intel should 'simply' do is to work with AMD, crazy but having an array of different hybrid layouts and implementations (2+1, 1+5, 6+8 etc) will spell nightmare for the operating system vendors and app engineers even GPUs will be affected. Is it time to update the X86 architecture? AMD is also to blame for not working on AVX512 (or is it plain marketing).

I'm not sure what my particular feelings about hybrid CPUs are. They tend to work well in the mobile space and even more recently in the GPU space (Nvidia). I guess it really comes down to how well the scheduler and software copes with cores of varying capabilities, I'm sure some software can be pushed to the lower power cores depending on the instruction sets those cores have. I'm guessing it all depends on how pared back those smaller cores are in terms of capabilities.

I guess time will tell, it's certainly interesting to see exactly how Intel / AMD / ARM / Apple will differ in their approaches.

[Tabbykatze]

"potential pitfalls of hybrid processors coming to Windows." why end with such a negative statement? what is the agenda?

Not an agenda, a general statement of fact as Windows wholly is not optimised for Hybrid architectures (yet). They're changing over 20 years of the current iteration of Windows (Win 10 still technically hails from Blackcomb in 2003 then became Vienna in 2006 and is a massive iterative improvement from Win 7 to Win 8 but still using the same base HAL). What this means is you are changing nearly 20 years of traditional operation of standard multi-cores (all being the same) to now suddenly have mismatch core configuration.

So what's your agenda?

Waiting for SkyN or whatever his name was to make an appearance.

[Iota]

What this means is you are changing nearly 20 years of traditional operation of standard multi-cores (all being the same) to now suddenly have mismatch core configuration.

Wouldn't it be interesting to have all the background Windows processes running on the low power cores, with the bulk of normal programs running on the larger cores.

Also if the quote is taken in full context instead of the cherry picked version, I would imagine the only agenda is one from Intel to make sure the transition is as smooth as possible while avoiding the potential pitfalls of hybrid processors. As you've mentioned, Windows isn't optimised for Hybrid CPU architectures (yet).

[3dcandy]

The scheduler has had a few changes over the last couple of years to help in this matter of course. The M1 chip is a pretty standard Arm big.little chip design and isn't really anything different than has been made before - what is different is it's being incorporated into a different type of system ie. laptop and desktop. The linux kernel and schedulers work well in this respect and that is basically how Apple and mobiles work so well....
One of the problems which will happen is people hanging on to old software especially games which won't like this approach and also a lot of different programming idea need to be implemented. Part of the Windows problem here (I cannot see many issues at all on Linux as the main flavours already work well with Arm big.little setups) is that differing core setups haven't been about before, and to be fair have not really been thought of

[iEimis]

The scheduler has had a few changes over the last couple of years to help in this matter of course. The M1 chip is a pretty standard Arm big.little chip design and isn't really anything different than has been made before - what is different is it's being incorporated into a different type of system ie. laptop and desktop. The linux kernel and schedulers work well in this respect and that is basically how Apple and mobiles work so well....
One of the problems which will happen is people hanging on to old software especially games which won't like this approach and also a lot of different programming idea need to be implemented. Part of the Windows problem here (I cannot see many issues at all on Linux as the main flavours already work well with Arm big.little setups) is that differing core setups haven't been about before, and to be fair have not really been thought of

Whilst I agree that it will take time, I think that it is achievable, for older software (not so sure), however, I've been using Rosetta 2 for a while now and what it emulates, it runs extremely well only losing around 30% of performance and still running faster than current intel offerings in the likes of Dell XPS. I think if Apple can continue executing on this and bringing continuous improvements in the next iterations they might quickly find themselves having a competitive advantage in the ultrabook segment at the very least.

[3dcandy]

Whilst I agree that it will take time, I think that it is achievable, for older software (not so sure), however, I've been using Rosetta 2 for a while now and what it emulates, it runs extremely well only losing around 30% of performance and still running faster than current intel offerings in the likes of Dell XPS. I think if Apple can continue executing on this and bringing continuous improvements in the next iterations they might quickly find themselves having a competitive advantage in the ultrabook segment at the very least.

You can't use that example as Apple has used several million transistors to accelerate that in the M1 processor - it's not a fair apples to apples comparison

[watercooled]

In response to the "we've had big.little for years on ARM" comments - that's not the concern here. Scheduling for different sizes of cores is challenging but quite doable with current techniques (obviously). The issue is the rumour of instruction incompatibility between the large and small cores - if that isn't handled correctly, it's not just a case of performance issues, you get system crashes. In all examples I can think of, the big and little cores have identical instruction support, and is why you can only pair certain cores together.

[iEimis]

You can't use that example as Apple has used several million transistors to accelerate that in the M1 processor - it's not a fair apples to apples comparison

True but what they achieved is pretty amazing, Windows ARM has been available for several years now and it's worse natively than Apple M1 running a VM instance on Rosetta and running another Windows VM inside that emulated VM instance which is quite funny, to say the least. Qualcomm just isn't there yet, but competition is healthy for everyone.

[3dcandy]

In response to the "we've had big.little for years on ARM" comments - that's not the concern here. Scheduling for different sizes of cores is challenging but quite doable with current techniques (obviously). The issue is the rumour of instruction incompatibility between the large and small cores - if that isn't handled correctly, it's not just a case of performance issues, you get system crashes. In all examples I can think of, the big and little cores have identical instruction support, and is why you can only pair certain cores together.

But this is why/where and how Windows is lagging behind. I'm not worried about it per se but I agree it means a total shift in how the scheduler will work. I'm sure we can all remember the scheduler issues when Ryzen first came out and this is another of those things. Intel *seem* to be trying to do something that isn't on the surface required or even asked for. If now is ever the time that Linux laptops should start to become appealing I can't when they will ever get that chance. I was out of Linux tweaking for a few years but have built myself a Xubuntu server for livestreaming. If I didn't need some Windows software I'd be very tempted to go all Linux right now

[3dcandy]

True but what they achieved is pretty amazing, Windows ARM has been available for several years now and it's worse natively than Apple M1 running a VM instance on Rosetta and running another Windows VM inside that emulated VM instance which is quite funny, to say the least. Qualcomm just isn't there yet, but competition is healthy for everyone.

Again I really can't see what all the fuss is about the M1. They have all the resources to do it, put a load of transistors into the design to hardware accelerate certain things and have perfect ties between the whole design process including the OS and everything. It's a decent chip with a decent amount of OS support - but Qualcomm etc. could do a pretty decent job if anybody else had requested something and in those volumes. Nobody mentions all the things they have taken out/removed like decent USB channels etc. to accelerate things and whilst in a thin and light laptop or the Mac Mini that is fine, but the vast majority of Windows users wouldn't accept those limitations let alone removing that hardware support would result in Windows not really running a lot of older software. I see that running Rosetta 2 is great, but it's not going to help the Microsoft users when older software runs poorly (like some MacOS stuff does now) because the vast majority of users don't want that. Windows ARM was never given the serious support or resources it needed to flourish, and users didn't really want it either so it just stagnated, along with the chip designs to help it. Apple can do both - support the hardware they made with an OS they design and code for, something that Windows will never be able to do, but just as big a problem is most Windows user (majority not the type of users on this website/forum) have no interest in a lot of things the M1 is designed to do

[iEimis]

Again I really can't see what all the fuss is about the M1. They have all the resources to do it, put a load of transistors into the design to hardware accelerate certain things and have perfect ties between the whole design process including the OS and everything. It's a decent chip with a decent amount of OS support - but Qualcomm etc. could do a pretty decent job if anybody else had requested something and in those volumes. Nobody mentions all the things they have taken out/removed like decent USB channels etc. to accelerate things and whilst in a thin and light laptop or the Mac Mini that is fine, but the vast majority of Windows users wouldn't accept those limitations let alone removing that hardware support would result in Windows not really running a lot of older software. I see that running Rosetta 2 is great, but it's not going to help the Microsoft users when older software runs poorly (like some MacOS stuff does now) because the vast majority of users don't want that. Windows ARM was never given the serious support or resources it needed to flourish, and users didn't really want it either so it just stagnated, along with the chip designs to help it. Apple can do both - support the hardware they made with an OS they design and code for, something that Windows will never be able to do, but just as big a problem is most Windows user (majority not the type of users on this website/forum) have no interest in a lot of things the M1 is designed to do

You put it perfectly yourself, for thin and light it's fine, in fact, better than fine really, for the first time in a long time you have a pretty capable mobile video editing machine (running Final Cut Pro X is the best of course due to optimisations). But Interesting to see where they take the development in terms of the silicon when iMac's and Mac Pro's are concerned. I guess we will see soon.

[3dcandy]

You put it perfectly yourself, for thin and light it's fine, in fact, better than fine really, for the first time in a long time you have a pretty capable mobile video editing machine (running Final Cut Pro X is the best of course due to optimisations). But Interesting to see where they take the development in terms of the silicon when iMac's and Mac Pro's are concerned. I guess we will see soon.

Indeed - but the vast majority of Windows users aren't bothered about that... and Windows is what we are talking about. Don't you think/agree/disagree that if Microsoft owned Intel they'd have done something like this years ago?