Question Alder Lake - Official Thread

Page 59 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

tamz_msc

Diamond Member
Jan 5, 2017
3,710
3,554
136
Okay I'm going to admit it. I think Ian was right. The thread director isn't keeping enough compute on Handbrake when I'm transcoding something. For example, right now as I type this it's encoding at 16fps, but when I switch apps and put Handbrake on top it increases to 44fps. Now I could understand keeping Handbrake on the E's if I was doing something that required the P's, like gaming or something, but I'm typing a post in Chrome! No reading to have 8 P's standing at the ready doing nothing.

And yes, I realize I can get this going correctly with process lasso but I shouldn't have to. If you're simply typing in Word or something there is no reason the bulk of the processing power can't go to the background Handbrake encoding automatically. Maybe if there were 16 E's or something this behavior wouldn't be so bad but right now it's kind of dumb.
Have you tried changing the power plan to Best Performance?
 

TESKATLIPOKA

Platinum Member
May 1, 2020
2,329
2,811
106
Intel has always stated that the E cores were for multithreading and the P cores were power hungry but best at single threading. So, no it is not at all surprising when you take a 125+ W chip, give it only 50 W, have it's P cores suck up the little power you gave it, and you are left with an underperforming chip in a multi-threaded application. At the very least, wait for the non-K versions (especially the mobile or even the T versions) to see how they work at lower powers.
I suspect that at 50 W, the E cores aren't getting 3 W each. At least in the higher power situations, the P cores are taking about 4.5x the power as the little cores. If that ratio stays true when you run down to 50 W, you are looking closer to 5 W each for the P cores and just over 1 W each for the E cores. That might help you get more accurate estimates than assuming 3 W each to all 16 cores.
Of course P-cores consume more power than E-cores, that's why I wrote 2.75-3W on average, because I don't know the exact amount for each core, but let's say your power consumption estimate for each core is correct.

i9-12900K's P-cores work at ~4.9GHz and E-cores work at ~3.7GHz with PL1=PL2=241W and generate 27780 points in CB R23

To score only 8872 in CB R23 by limiting It to 50W would mean that P-cores are working at ~1568Mhz and E-cores at 1184MHz.(I downscaled the clock for both by the same ratio)
A single P-core at 1.6GHz shouldn't consume 5W.
Even 1W for E-core at 1.2GHz is too much. Based on what Raichu posted, I calculated here that a single E-core with 0.8V working at 2GHz should consume ~0.9W. At 1.2GHz It should consume even less power.
What I want to say is that the score is quite low for 50W.
 
Last edited:

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,686
136
I found something very interesting.
i9-12900K limited to 50W for both PL1 and PL2 provides only 8872 points in CB R23 and the power consumption is 120W.
If you check Intel Core i9-12900K E-Cores Only Performance Review then you will find out that i9-12900K with only E-cores enabled provides 10366 points in CB R23 and consumes 118W.
You gain ~17% higher score and power consumption is actually 2W lower.
Just E-cores are better than a combination of P+E cores not just in efficiency but also in performance at 50W power limit based on TPU reviews.
Is that really not surprising?
There were several users here who cautioned against taking the results from the latest TPU review at face value.

Here's 12700K scoring 12k+ in CB23 with PL1=PL2=50W.
CB23-50W.jpg
 
  • Like
Reactions: lightmanek

TESKATLIPOKA

Platinum Member
May 1, 2020
2,329
2,811
106
I think in that TPU review the PL1=PL2 was limited to different values, but voltage stayed the same or something like that.
This would explain the high power consumption for such a low score.
 
Last edited:

Hulk

Diamond Member
Oct 9, 1999
4,191
1,975
136
Have you tried changing the power plan to Best Performance?

Yes I did. No difference. Power plans really only change how fast various components ramp up not how CPU's are utilized. The easiest way to to adjust things is with process lasso. It's pretty easy actually and I can tailor to my workflow. For example, I have DxORaw set to use 6 P's regardless. So when I'm editing in Photoshop I have 2P's and 4E's, which is plenty. Meanwhile DxORaw is processing the RAW images with the 6P's.

Honestly the 4E's processing RAW files in DxOPure Raw can almost keep up with me as I edit in PS. 4P's would probably be enough. By "keeping up" I mean the time for DxO PureRaw to process a RAW is about the same as me to edit with PS. 8E's would probably have done it. I shouldn't have cheaped out and gone for the 12900K!
 
  • Like
Reactions: lightmanek

Asterox

Golden Member
May 15, 2012
1,026
1,775
136

Chinese 12400F review. Performance looks good, manages to be faster than stock 5600X.

Well, this is not news that is faster in Cinebench and CPU-Z.


I think the moon would collapse, in a situation where someone wouldn't use Cinebench and CPU-Z.:mask:
 

tamz_msc

Diamond Member
Jan 5, 2017
3,710
3,554
136
Yes I did. No difference.
That's strange. The Anandtech review specifically mentions that the issue with threads being assigned to E cores when the app is not in focus can be addressed in the following three ways:

  1. Running dual monitors stops it
  2. Changing Windows Power Plan from Balanced to High Performance stops it
  3. There’s an option in the BIOS that, when enabled, means the Scroll Lock can be used to disable/park the E-cores, meaning nothing will be scheduled on them when the Scroll Lock is active.
 

Hulk

Diamond Member
Oct 9, 1999
4,191
1,975
136
That's strange. The Anandtech review specifically mentions that the issue with threads being assigned to E cores when the app is not in focus can be addressed in the following three ways:



I had a hard time changing the power plan. Going straight to "Power Plan" didn't seem to work since when you select it "Apply" remains greyed out. Hit okay and go back in and it's still showing "Balanced." I found the command to change it through CMD window and that seemed to work. But still a the app goes to the E cores when it goes to the background. So far the easiest, most reliable way to get the behavior I want is to simply set affinities in Process Lasso. They work and they stick. There are only a couple of apps I need to keep the P's on so I just set them up in Process Lasso.
 

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,686
136
That's strange. The Anandtech review specifically mentions that the issue with threads being assigned to E cores when the app is not in focus can be addressed in the following three ways:
On Win 10 moving to High Performance doesn't help, and the situation can actually get worse depending on the video encoder you're using. For Handbrake, on default settings (thread priority set to "Bellow Normal"), video encoding may end up on E-core only irrespective on whether the application is on the foreground or not. Handbrake also has settings for thread priority (the same as the ones available through Task Manager).

Here's what happens to the same video conversion depending on the thread priority setting. Remember, this is Win10, and the last case illustrates what happens when another app is brought to the foreground:

handbrake-core-alloc.png
 

Hulk

Diamond Member
Oct 9, 1999
4,191
1,975
136
Does switching to Win11 fix the problem?

I'm on Windows 11.

To put it bluntly the scheduler or whatever is making CPU loading decisions is behaving stupidly.
For example, I'm processing raw images with DxO PureRaw, it's a very compute intensive process. At the same time I'm editing those processed raw images in PS. I can generally post process an image in PS in 60 seconds or so. But while PS is on top ONLY the E's are working in PureRaw and it takes them 164 seconds to process an image. Even if I'm doing nothing in PS all 8 P's will sit parked while only the E's work in Pureraw. And of course I'm waiting for the next image to process.

Using process Lasso I set PureRaw to always use 4P+4E, which brings down image processing time to 41 seconds per image. By the time I'm editing an image PureRaw has finished processing so all compute is available in PS and even if it wasn't 4P's is generally sufficient for near realtime performance in PS.

That's my workaround for a couple of my situations where Windows 11 isn't behaving as I'd like.

The Windows 11 Scheduler behavior I don't understand is why when the application that is currently on top is idle or nearly so the rest of the compute doesn't go to background tasks? Or even 80% so the system can feel responsive to the user when they do something during that fraction of a second where the compute resources switch back to the foreground application?

It's seems stupid the way it is behaving but I'm sure I'm missing something and I'm the one who is stupid because there is no way that teams of engineers and beta testers who are all magnitudes of order smarter than me could have missed this.
 
Jul 27, 2020
15,749
9,816
106
It's seems stupid the way it is behaving but I'm sure I'm missing something and I'm the one who is stupid because there is no way that teams of engineers and beta testers who are all magnitudes of order smarter than me could have missed this.
The same teams of engineers and beta testers who were responsible for the AMD cache latency issues in Windows 11 RTM?
 

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,686
136
Not sure how these would look in Win10, but performance seems perfectly fine to me for 1TB Firecuda 530:
Supposedly it's fixed.
My understanding is that the issue affects only certain systems, otherwise it would have been easier to spot in time. I simply don't feel like beta testing for MS, little to win on my side especially considering I need to use this machine for work. I purposely built it and installed the OS on Friday evening so that by Sunday I could decide whether the new machine stays in place or I bring back the old system. I'll reconsider a clean Win11 install at the end of the month. The E-core behavior is weird but it won't affect my workflow.

The Windows 11 Scheduler behavior I don't understand is why when the application that is currently on top is idle or nearly so the rest of the compute doesn't go to background tasks? Or even 80% so the system can feel responsive to the user when they do something during that fraction of a second where the compute resources switch back to the foreground application?
The scheduler behavior is what happens when a major software component development is linked to an arbitrary launch such as Nov 2021 for Alder Lake. It was good enough for review benchmarks, so they shipped. As I said before, I hope Intel puts in serious effort and fixes this for the mobile launch. I plan to disable the E-cores as soon as firmware allows me, but at the same time I also plan to enable them back once team blue figures out how to make E-cores efficient to work with.

Meanwhile Intel managed to bring the DRM situation under control, so I guess that's something.
 

Hulk

Diamond Member
Oct 9, 1999
4,191
1,975
136
My understanding is that the issue affects only certain systems, otherwise it would have been easier to spot in time. I simply don't feel like beta testing for MS, little to win on my side especially considering I need to use this machine for work. I purposely built it and installed the OS on Friday evening so that by Sunday I could decide whether the new machine stays in place or I bring back the old system. I'll reconsider a clean Win11 install at the end of the month. The E-core behavior is weird but it won't affect my workflow.

I was wary of Windows 11 as well. Not so much from the performance quirks but more the messing around with the GUI that MS can't seem to resist. Little things like you can't right-click the taskbar to pull up task manager, you need to right-click over the start icon to do that. I adapted in a week or so and now I don't notice the 11 vs 10 GUI differences so much

Yes, you are right regarding getting it out with good benchmark scores was the priority. As long as the benchmark application is in the foreground (or any app for that matter) then it will receive the full compute of the CPU. I think the fact that the E's and only the E's automatically go to background tasks is more pronounced with the parts that only have 4 E cores. If I had a 12900K and Handbrake or DxO PureRaw was running solely on the E's when running in the background the performance would probably have been adequate for me and made this non-issue for the most part. If Raptor goes to 16 E's then I think for me this issue goes away.

Similarly for 2+8 mobile the current behavior might actually be preferable. Using my 12700K as an example I find that 4E's are roughly 19% "stronger" than 1 P. So, for ultra mobile 2+8, 8 E's should be more powerful than 2 P's and do a good job with background tasks. Since only 2 P's are available they would need to stay with the foreground tasks. The other mobile configurations might get a little dicey though with this issue...
 

lobz

Platinum Member
Feb 10, 2017
2,057
2,856
136
That is truly horrific behavior for a dedicated hardware assistant with no easy / user friendly way to configure it. Suggesting that every ADL customer is their own system administrator ia also just totally tone deaf, but hey...
Let's see what becomes of this in a generation or two. For example, DLSS too has started out as an unintended blur filter after all.
 
  • Like
Reactions: Thunder 57

lobz

Platinum Member
Feb 10, 2017
2,057
2,856
136
Supposedly it's fixed.
Every let's do a clear install every week guy should switch to win 11, they make the maturing process a lot faster. Anyone who either wants to Install windows ONLY when switching computers or major parts (me), or anyone who actually needs to do work on their machine (me at work), should avoid newer windows versions like boiling lava 😂
That is the experience of my last 20 years as a PC enthusiast. Heck, I even held out on Win 7 till last Spring, when I upgraded to 9900K. Same with XP back in the day... I even endured random crashes on 98 till SP1 came out and I could switch off that stupid mouse accel under XP.
 
  • Like
Reactions: igor_kavinski