Intel 12th Gen Alder Lake T-Series 35W CPUs Reportedly Hit 4.9 GHz

[Vishera]

10 minutes ago, Mark Kaine said:

but I do have the feeling those cpus are going to suck for gaming in general,

Gaming is a mixed workload that relies on both single core performance and multi-threaded performance.

So the little cores will have significantly less single core performance compared to "normal" cores.

 

I expect Intel to have a solution to that problem (they could disable those cores in gaming workloads)

[Mark Kaine]

3 minutes ago, Vishera said:

I expect Intel to have a solution to that problem (they could disable those cores in gaming workloads)

I was thinking about that possibility too... aside from that being probably complex ,that would mean a 6c/12t 12th gen intel basically runs on how many cores while gaming? That just seems like regress and I can easily see a situation where games need to specifically optimized for this... resulting in chaos, incompatibilities,  low performance,  suffering,  blood sweat and tears ...

 

Maybe these cpus are actually good , just doesn't seem like gaming had any priority or even consideration while designing them. of course they will claim the opposite,  but im skeptically curious...

 

 

 

[porina]

35 minutes ago, Vishera said:

I expect Intel to have a solution to that problem

From what we've been shown already, this is a combination of Intel's Thread Director (provided in CPU) and OS support. Between them they can work out what makes sense to put on P cores, and what can go on E cores. It is a solution that doesn't require software support, although software that can understand Alder Lake may be able to do something smarter if needed. The writer of Prime95 is arranging pre-release access to try and get it working well in preparation. Up to now it is tuned to assume all cores are equal, since they have a degree of needing to sync up, which will no longer be the case. Of course, as something new, we will have to wait and see how it really works in practice. Win11 will be a requirement and that in itself may put people off. Win10 will not understand the difference between the cores so performance may less unpredictable.

 

We can also kind of see this as similar to SMT in terms of putting the right code on the right threads. In the worst case you might get 50% performance if you put two threads on the same core that should have gone on separate cores. Generally speaking, with SMT enabled you should not see any major drop in performance compared to SMT disabled. Windows has got better over time I think we're at that point now. Note that doesn't mean no negative difference ever. There will be a small overhead from SMT existing, and it also takes up a little power. In older versions of Windows, for workloads I characterised, I saw SMT give between -10% to +50% improvement. The low end has gone up to -2% or so now as it does things better. On average, the benefits outweigh the negatives, but in a power constrained situation that doesn't benefit from SMT, you could see reduced performance that way.

 

The same test should be applied to hybrid CPUs going forward. The addition of the E cores should not result in scenarios where you get significantly lower performance compared to not having them. This will be an easy test for reviewers once it is available. I'm sure many will also explore how bad an idea it might be to run Win10 with it.

[leadeater]

2 hours ago, Mark Kaine said:

I was thinking about that possibility too... aside from that being probably complex ,that would mean a 6c/12t 12th gen intel basically runs on how many cores while gaming? That just seems like regress and I can easily see a situation where games need to specifically optimized for this... resulting in chaos, incompatibilities,  low performance,  suffering,  blood sweat and tears ...

 

Maybe these cpus are actually good , just doesn't seem like gaming had any priority or even consideration while designing them. of course they will claim the opposite,  but im skeptically curious...

I don't think it's going to be too much of a problem on the desktop CPUs with 4 or more enabled P cores as threads by default start on P cores, unless under conditions where that cannot be meet. I think laptops will have the hardest time as many CPUs are only going to have 2 P cores which isn't a lot when a load of threads and handles want access to or would be better if run on P cores.

[Mark Kaine]

12 hours ago, leadeater said:

I don't think it's going to be too much of a problem on the desktop CPUs with 4 or more enabled P cores as threads by default start on P cores, unless under conditions where that cannot be meet. I think laptops will have the hardest time as many CPUs are only going to have 2 P cores which isn't a lot when a load of threads and handles want access to or would be better if run on P cores.

I didn't mean it would totally suck, more like potentially not a big improvement over previous gens... its something "new" i guess and i applaud them for doing something different - but i cannot ignore how disconnected the company seems to be either.  Remember the whole "back to you, Steve" PR farce recently, like what the heck was that? Lol

 

We will see , but as i said i can definitely see issues with this approach,  these things aren't always as easy as they seem... video games specifically are weird and kinda unpredictable regarding compatibility etc.

 

 

[Rauten]

I'm actually of the opinion that, if they haven't somehow screwed up the pooch, these could be awesome for budget-conscious gaming builds

.

Not just a 6 core, but a 6 core with a couple "weak" cores sprinkled in that can take care of that Discord call and those RAM-munching Chrome tabs.
Even more noticeable for the i3s - you're only gonna have 4 decent cores, but (OS scheduling permitting) they will be laser-focused on your game.

 

I'm very interested to see how these new cpus work out.

[Maticks]

honestly don't know why they can't have little cores, big cores and one extreme core that is 6Ghz but only one core. throw single threat higher CPU usage like some poorly optimised games onto that core. at least its more likely to shine vs any other processor be it old intel ones or new amd chips.

[leadeater]

43 minutes ago, Maticks said:

honestly don't know why they can't have little cores, big cores and one extreme core that is 6Ghz but only one core. throw single threat higher CPU usage like some poorly optimised games onto that core. at least its more likely to shine vs any other processor be it old intel ones or new amd chips.

Probably because it'll really only benefit older games using DX11. Not much coming out now days is dependent on a single core or 2 thread (SMT). It's basically a solution to a problem that exists in the past, although it would give some nice CB 1T scores.

[porina]

1 hour ago, Maticks said:

one extreme core that is 6Ghz but only one core

If you literally mean 6 GHz, the danger is we'll end up with something like Pentium 4 again. It'll have to be designed for high clock, but it doesn't necessarily follow it'll be high performance.

 

CPU cores from both sides are already pretty balanced on a performance basis. If it was easy to make a higher performing core without breaking either the power budget or cost budget, they would. Basically it'll be a lot of pain to deliver something that would only be a niche benefit.

[AnonymousGuy]

On 9/19/2021 at 4:54 AM, Mark Kaine said:

Yeh, but the post you quoted said specifically for gaming... tbh I dont *really* know what's the point of "little" cores but I do have the feeling those cpus are going to suck for gaming in general,  coupled with weird DDR5 shenanigans (latency issues) its probably gonna suck even more. We will see when there a proper benchmarks , but this is what it looks like to me currently. 

 

It's going to be transparent to you because the OS and microcode are going to be scheduling tasks anyways. The idea is that minor tasks (background processes, light game threads like sound/network) get shoveled to atom where it keeps the Core from getting interrupted and the atom cores work far more efficiently than Core does.  Roughly each Atom core counts for 60% of a Core, and yeah Intel did explore just putting 12 Cores together instead of doing 8+8.

 

Long term the goal is laptops and datacenters.  Intel is trying to get ahead of ARM taking over servers with 9000 core boxes, so they're going to make their own 256 core atom parts.  It's obviously not for all customers though but certain shit like networking scale up really well to high core counts.   Laptops: well there's obvious advantages to using higher performance/watt Atom cores than Cores.