Discussion AMD 3600x and Windows 10 Affinity Settings

dmitche31958

Reputable
Aug 13, 2019
43
6
4,545
Can anybody explain the current thought process behind AMD 3000 CPU and Windows 10 Affinity options? I previously had a FX-8370 8-core and upgraded to the 3600X with Vengance 3600 memory. In an attempt to keep the CPU cooler while encoding video I play around with the affinity settings. Under the old CPU it was a pretty straight forward process. If you lower the number of threads the amount of CPU usage would drop as well as the temperature.

Under the 3600x this isn't true. It's very crazy and doesn't make much sense with some one like myself without additional knowledge of how Windows is deciding how to handle the CPU.
For example.

Using the Task Manage and de-selecting under the affinity option 'CPUs 6 and 10 I see using CPUID Hardware Monitor that the wattage decrease for "Core 5" drops to what I would say as "not used".

Turning off what Task Manager calls CPUs 4,5, 6 will turn off cores 2,4 and reduce wattage by 3 watts to core 3. Okay, pretty close to one to one action to effect.

Turning off CPU 0,1, and 2 shows cores 0,3, 4, and 5 as not used and core 1 down to 0-2W averaging 1.7 W (from 13-14 at full power. Now THIS IS weird!!!!
Also, turning off these 4 outof 12 'CPUS' reduces the CPU usage for the process from 90% down to 16%.

As you can see just changing the affinity options under Task Manager isn't as simple as removing threads being used from a process.

With cores/CPUS 0, 1, and 2 disabled the CPU package is still running at 71C, only 9C cooler than running full out, all cores at about 11W, 65W for the core and a total of 92W total for the package.

In addition, the temperature changes are just as wacky. I haven't recorded a table of changes for that yet but the changes in CPU package is even weirder. Removing cores by the Affinity option ACUTALLY INCREASED CPU temperatures!!! I initially decided to turn off cores 0-4

Turning of 'CPUS' 8,9 and 10 shows cores 4 & 5 no longer being used and core 5's power cut in half. Temperature wise the CPU runs at 77C, a drop of only 3 degrees Celcius. CPU usage is 66%. Hmm, a 25% cut in threads but produces a larger drop in CPU usage but without much change in temperature. Strange.

I would desperately like to understand AMD and Windows attempt to handle processes and CPU cores.

Please, don't respond if you aren't knowledgeable about the process. I don't need to check my memory clocking or look for a virus.

Adddum; Turning off cores 0 and 3 reduces the CPU usage to 66%, same as turning off 8,9, and 10 and with ZERO change in temperature.
 

dmitche31958

Reputable
Aug 13, 2019
43
6
4,545
Addendum: Turning off cores 0,1,3,5,and 8 reduces CPU usage for the process to 8%! and a temperature of 66C!!!
Reducing 1/2 the threads causes a 90% drop in thread processing(CPU) but the temperature remains extremely high for only 8% CPU usage!!
 

DMAN999

Honorable
Ambassador
What are your full PC specifications, including full make and model of every component.
Also what CPU cooler are you using and what case and with how many case fans ?
What is the temperature in the room where the PC is being Used ?

I have a 3700x and my CPU idels at 32-34c and maxes out at 60-62c after hours of gaming.
That is with all cores active and SMT and PBO enabled.
My room temp is 23c and I have a 5 case fans (3 in and 2 out).
 
Last edited:
...

In addition, the temperature changes are just as wacky. I haven't recorded a table of changes for that yet but the changes in CPU package is even weirder. Removing cores by the Affinity option ACUTALLY INCREASED CPU temperatures!!! I initially decided to turn off cores 0-4

...

There is not just one temperature sensor for Ry3k... there are several with at least one for each core. So, with that in mind, I suspect what's happening is as you remove cores from the available pool for the process the scheduler loads up the remaining cores more heavily since it doesn't have other cores to move the work around to anymore.

The temperature increase you see is the effect on the remaining core(s) as they have to work harder. This would make sense if the temperature reported is the highest temperature of all the sensors, which is what I've understood to be the case.

EDIT: BTW... as is much bemoaned by the tech illuminati (PC master race, whoever) AMD is extremely tight with tech data of just what's going on in Ryzen processors. So most of what's being passed around is what they've what they've learned by testing and putting together as much information as is possible from cause/effect relationship analysis along with what little technical detail they can get from the source. Point being: there's not many who actually 'know' a lot about these things, and those who do are locked up with NDA's so won't divulge.
 
Last edited:
  • Like
Reactions: DMAN999

dmitche31958

Reputable
Aug 13, 2019
43
6
4,545
Thanks Dman.
Amd 3600X
Vengence 3600
ASUS ROG X570 mother board
PSU Corsair 650
LOL. HD 6700 video card. My card blew out with my previous MOBO went bad so I'm using my fall back.
Full super tower. 5 Case fans. I run them at 100%
960 EVO 500GB SSD (OS,apps, and a few games)
WD Black 1TB ( data drive)
WD Red 3 TB video archive ( data )

Idling around 38-40C as my room isn't air conditioned and ambient temperature isn't as cool as most peoples.:) it's about 80F today under the fan when I ran those tests.:)
 
  • Like
Reactions: DMAN999

dmitche31958

Reputable
Aug 13, 2019
43
6
4,545
There is not just one temperature sensor for Ry3k... there are several with at least one for each core. So, with that in mind, I suspect what's happening is as you remove cores from the available pool for the process the scheduler loads up the remaining cores more heavily since it doesn't have other cores to move the work around to anymore.

The temperature increase you see is the effect on the remaining core(s) as they have to work harder. This would make sense if the temperature reported is the highest temperature of all the sensors, which is what I've understood to be the case.
I don't see that happening. I thought of that myself. That Windows/AMD thread logic is trying to apply the process to the "best thread" but it doesn't look that way as far as output. There is a problem that I think AMD and Microsoft need to address if this is the case. If I'm specifically telling Windows what threads to use or not use and it is ignoring it and even making things worse they need to acknowledge this and explain why they are making things worse.

But watching the amount of power that is being used I don't see an increase in any cores when I remove a core from the process. I only see either a 25% reduction, from an average of 13W to 9W and a reduction to less than 1W.

Also, I know that the Turbo or 'Boost' is causing the issues. Or I'm 99% sure of this. If I leave Affinity alone and change the Maximum porcessor state to 99% under Power options. Advanced settings I can see that running at 90% CPU reduces the temperature from 87C down to 56C. A 1% reduction in CPU also causes the wattage per core to be cut in half, from roughly 12.x/13 down to 6.xx and total wattage to 35W for the cores.
 
Last edited:
I don't see that happening. I thought of that myself. That Windows/AMD thread logic is trying to apply the process to the "best thread" but it doesn't look that way as far as output. There is a problem that I think AMD and Microsoft need to address if this is the case. If I'm specifically telling Windows what threads to use or not use and it is ignoring it and even making things worse they need to acknowledge this and explain why they are making things worse.

But watching the amount of power that is being used I don't see an increase in any cores when I remove a core from the process. I only see either a 25% reduction, from an average of 13W to 9W and a reduction to less than 1W.
What is the problem? that you can't manage the threads as well as their scheduler? I don't think they'd agree.
 

dmitche31958

Reputable
Aug 13, 2019
43
6
4,545
What is the problem? that you can't manage the threads as well as their scheduler? I don't think they'd agree.
Well, you don't read and think then. Read the initial post and say that any of that makes sense. Then repeat your answer to yourself, rinse, and repeat. LOL

And did I not say, don't respond if you don't know anything on the subject?
 

Latest posts