Question Alder Lake - Official Thread

Page 100 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

dullard

Elite Member
May 21, 2001
24,998
3,327
126
The fact that Alder Lake has an exotic core layout and a thread director scheme that behaves peculiarly might be responsible for the problem.
I would first swap mice. It could be a bad mouse. Then I'd reinstall drivers. The mouse driver and GPU drivers can do this and uncommonly other drivers could be causing a lag like that. Then I'd check Windows power saving settings. Those all are more likely causes of mouse problems than the way Alder Lake is built.
 

Exist50

Platinum Member
Aug 18, 2016
2,445
3,043
136
Could be an interconnect issue. Skylake CPUs didn't have the same interconnect layout, and if user input requests get constantly bounced between core clusters, there could be some lag.
Any interconnect latency would be well below human perception. And if that were actually a problem, would be noticed by more than one person.

The explanation is far more likely to be a wonky software installation than anything to do with Alder Lake, hybrid or otherwise.
 
  • Like
Reactions: Zucker2k

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,483
14,434
136
  • Like
Reactions: Drazick

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,483
14,434
136
Well, that machine was off. I turned it on and was trying to type in the search window, but it was like no keyboard. 10 seconds later it comes back. I can't find "Intel Rapid Storage" device, but it needed a windows update, so I will look again after that finishes. Thanks, I think I just verified it again.
OK, after hours of updating drivers. I MAY have fixed the problem. I can say its only windows 11 and Alderlake, but I cannot say what it is exactly.

Kind of like the AMD USB thing...... They now share something !
 

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,686
136
The dreaded Alder Lake USB Bug! Discovered right here on Anandtech forums.
Funny how in just 2-3 thread pages we've gone from "Thread Director may have hardware bugs" to "updating drivers solves issue, wasn't about Thread Director anyway".

People who promoted this FUD about broken hardware scheduling should at least eat some crow and start learning how to deal with anecdotal evidence.
 

JoeRambo

Golden Member
Jun 13, 2013
1,814
2,105
136
I haven't tested recently, but with 8 thread limited CB23 i was getting ~12-13k with HT enabled and 15-16k with HT disabled. No idea if this was fixed and who is/was to blame.

And scheduling looked completely wrong too, workload jumping all over the cores.
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,483
14,434
136
Funny how in just 2-3 thread pages we've gone from "Thread Director may have hardware bugs" to "updating drivers solves issue, wasn't about Thread Director anyway".

People who promoted this FUD about broken hardware scheduling should at least eat some crow and start learning how to deal with anecdotal evidence.
You did notice that there is only ONE post that says the issue MIGHT be solved by updating drivers. I only was being honest, as I did not see the issue for about 10 minutes after the last bunch of updates and driver updates. Don't be jumping to conclusions on my part, when I said MIGHT, I mean it. Trying not to be a negative impact, but real feedback. I have just never seen anything like this before.
 
  • Like
Reactions: Drazick

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,686
136
I haven't tested recently, but with 8 thread limited CB23 i was getting ~12-13k with HT enabled and 15-16k with HT disabled. No idea if this was fixed and who is/was to blame.

And scheduling looked completely wrong too, workload jumping all over the cores.
That's still happening on Win10, and from what I'm seeing on my machine there's a lot of thread movement on two of the P-cores, with other cores seeing less hopping (but still present).

I think an important point is that supposedly this issue is not a problem when E-Cores are enabled, just when only P-Cores are present.
Here's screenshots to illustrate what @JoeRambo is talking about. First Task Manager crop is 8+0/16t, the one bellow is 8+4/20t. Notice how threads no longer hop between logical cores.

cb23-HT-hopping.png
 

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,686
136
You did notice that there is only ONE post that says the issue MIGHT be solved by updating drivers. I only was being honest, as I did not see the issue for about 10 minutes after the last bunch of updates and driver updates. Don't be jumping to conclusions on my part, when I said MIGHT, I mean it. Trying not to be a negative impact, but real feedback. I have just never seen anything like this before.
You're not the one spreading the FUD in this thread, my issue is not with you reporting the problem or a POSSIBLE & unconfirmed solution.

The problem is users like @igor_kavinski who takes some anecdotal example and pushes for an unfounded conclusion like "ADL hardware scheduling is broken, cannot get fixed". As an engineer with decades of experience you are probably even more familiar with the pitfalls of pinning the fault of a malfunction on some component of a system while multiple components have been swapped/upgraded. Just for the perspective it brings, take a look at this simple "Win 11 mouse lag" search. There's all kinds of reports of mouse lagging in Win11 on hardware that did not lag in Win10 (on an AMD APU btw), there's reports of Win11 mouse lagging with HDR enabled.
 
Jul 27, 2020
15,759
9,822
106
Here's screenshots to illustrate what @JoeRambo is talking about. First Task Manager crop is 8+0/16t, the one bellow is 8+4/20t. Notice how threads no longer hop between logical cores.

View attachment 59311
Super interesting. Anyone notice how with E-cores disabled, the workload seems to be taxing 10 cores but with E-cores enabled, only 8 cores are handling the brunt of the workload?
 

JoeRambo

Golden Member
Jun 13, 2013
1,814
2,105
136
Super interesting. Anyone notice how with E-cores disabled, the workload seems to be taxing 10 cores but with E-cores enabled, only 8 cores are handling the brunt of the workload?

It's 8 thread sized workload, so 8 physical cores handling the load is the optimal behavior for this workload. And esp important is the sticking of same thread to same core it was scheduled before to keep L2 caches hot.

With E disabled, the workload is the same and after adding graphs they would be in 55-60% region ( at least that is what those % utilization tools show ), but the locality is gone and random threads are getting loaded, even if their HT sibling thread is already loaded. Horrible situation that results is 20% loss of performance in CB23 scenario and even hurts short running ST load of GB5.
Most likely it is present everywhere, cause i have noticed it when my specific workload showed strange degradation and performance was restored after disabling HT.
 

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,686
136
@coercitiv Comments please? There IS some weirdness going on with ADL.
Comments?! All of this and much more was discussed by @JoeRambo , @Hulk and myself on various occasions since we bought and tested out ADL systems. This ain't new at all for people following this thread since Dec 2021. More than that, I'm one of the more vocal skeptics when it comes to Intel's hybrid approach for the desktop (there's a reason I joke about "Thread Dictator"), so take that as a further indicator of how unfounded or alternatively blown out of proportions your comments about Thread Director are.

I guess I'll give you comments:
  • first you claimed there's an issue with E-cores enabled, that users are losing performance because of them
  • second you claim there's an issue with E-cores DISABLED, which shows this ain't about ITD anymore (you're just fishing for proof at this point)
  • so far you have made no differentiation between Win10 / Win11 / Linux, which figures since any OS doing significantly better than another would mean the issue(s) are software bound
  • you have yet to show a single reproducible scenario in which the average user is negatively impacted by ITD under Win11, the recommended OS for Alder Lake
If you want to criticize Intel's approach with hybrids on the consumer side, please go ahead and document your case, but that means more than anecdotal examples or extrapolating a Win10 finding to Win11 due to lack of testing.
 

Exist50

Platinum Member
Aug 18, 2016
2,445
3,043
136
Hyperthreading has always slightly harmed single thread performance. An inherent result of its static resource partitioning. If it weren't already developed and so useful for server throughput, it probably wouldn't exist for client chips now that hybrid is a thing.
 
  • Like
Reactions: coercitiv

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,686
136
I did a quick install of Win11 on a spare drive, thread migration behaves differently even with E-cores enabled. Bellow you can compare CB23 running 8t on 8+4 12c/12t CPU.

Win 11 on the left, Win10 on the right.
cb23-12c12t-win11-vs-win10.png

Score is obviously lower on Win11. One possible explanation, if this behavior intentional, would be that the OS is attempting to leave some P core resources for priority tasks (such as UI). Doesn't make much sense to me, but that's what we get for now.

Off-topic: Win11 isn't impressing me at all at first sight. I actually like the new presentation and I'm happy with the UI fixes, but performance wise it presented me with a poor first impression. It felt more sluggish than my Win10 install, to the point that I went back to Win10 to check whether I had disabled some UI animations and/or effects. Everything was default though. I'm starting to think that MS moving to a new Win version was a blessing in disguise, as otherwise I would have upgraded by now.
 
  • Like
Reactions: igor_kavinski

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
Any interconnect latency would be well below human perception.

If it's one hop, then yes. If not . . .

And if that were actually a problem, would be noticed by more than one person.

Well it kinda was.

The explanation is far more likely to be a wonky software installation than anything to do with Alder Lake, hybrid or otherwise.

Looks like it's more of an OS + platform issue, due to the way Windows is handling usb mouse/keyboard drivers on those systems.

Funny how in just 2-3 thread pages we've gone from "Thread Director may have hardware bugs" to "updating drivers solves issue, wasn't about Thread Director anyway".

People who promoted this FUD about broken hardware scheduling should at least eat some crow and start learning how to deal with anecdotal evidence.

Methinks thou dost protest too much. It wasn't going to be Thread Director in Mark's case since he leaves the e-cores disabled anyway. And it certainly wasn't going to be the e-cores themselves causing mouse slowdown since Core2 systems could drive USB mice just fine. Most of the commentary about Thread Director being problematic relate to matters having nothing to do with mouse/keyboard functionality.