What do you think of CPU Unparking? Real or gimmick?

Berryracer

Platinum Member
Oct 4, 2006
2,779
1
81
Do you think there are any benefits to CPU unparking or is it a gimmick?

I have been using the below tweak for 3 months but I don't feel I am missing anything without doing it after a format.

Can someone shed some light please:

If you are an owner of a new multi-core Intel CPU for example Intel core I7 and Windows 7 or Windows Server 2008 OS, you already might notice that some of the cores in your resource monitor are marked as parked. This is a new feature of windows 7/2008 operating system that is made to balance energy consumption by your CPU. So let’s say if you are performing some tasks that do not consume a lot of CPU power, all the cores that are parked will remain in that state.

However, if you are running something that requires a lot of CPU power, all the cores, which were previously parked, will be placed in the active state (unparked) to perform the task. And after it’s done, they will be parked again.


All in all, this is a nice feature to save the energy. But if you decide to keep all of your cores active at all times, there is no way to disable CPU parking from the user interface or by running command prompt.
trans.gif



This utility will allow you to easily enable or disable core parking for your CPU. The first thing that you should do is to go to a resource manager to check if you have parked cores. The reason of doing so is that core parking is not enabled for all the multi-core CPU’s. For example, if you have Intel Q9550 Quad Core CPU you may not see any parked cores at all.


resoure-manager-u.png



Once you have started “Manage Parked CPU Utility,” you have to press “Check status button” to allow program to search registry for the keys and values that are responsible for the “core parking”.

manage-core-parking-util-1.png


manage-core-parking-util-2.png



manage-core-parking-util-3.png



After search has finished, you will see N number of rows in the list view control along with the Status that will describe if the value retrieved from the registry indicates that your core is parked. But as I said earlier, core parking is not enabled for all the CPU’ s even though registry value may indicate that cores are parked. From now you can press “Park All” or “Unpark All” buttons to enable or disable parking. Also, you can change a single row.


Note that for the change to take effect you will have to reboot your machine. Also Utility should run with Administrator priveleges in order to be able to access registry. This utility does not require installation.


Download CPU Unpark Utility or you can download this update from Microsoft that selectively disables the Core Parking feature in Windows 7


Source: Coder Bag
 

Maximilian

Lifer
Feb 8, 2004
12,603
9
81
I dont think there is any benefit to messing with core parking, nice utility though. On my 3930k with windows 8 it dosent look like its parking anything according to resource monitor :eek: But the utility returned two values stating "parked" so i guess core parking is working correctly.
 

Nothinman

Elite Member
Sep 14, 2001
30,672
0
0
I think it's the responsibility of the OS and not something worth worrying about.
 

ViRGE

Elite Member, Moderator Emeritus
Oct 9, 1999
31,516
167
106
I think it's the responsibility of the OS and not something worth worrying about.
Agreed. This is why we have OSes, so that they can handle mundane things like this. It's best to leave the OS alone; it knows what it's doing.
 

imagoon

Diamond Member
Feb 19, 2003
5,199
0
0
As I recall, Parking is an OS function and the number of cycles to park -> unpark varies by CPU. In the Intel world I don't see a lot of complaints and most of the people that do it only say the machine "feels snappier" which could honestly just be "I did something so it must be better" syndrome. I know on the AMD bulldozer there is some shared cpu structures and parking one of the 2 cores can cause stalls when CPU 0 is running and CPU 1 is parked and CPU 0 uses a shared resource that 1 has parked. It makes CPU 0 wait until the CPU comes back online. I consider that something that needs to be altered in the OS scheduler though.
 

hamunaptra

Senior member
May 24, 2005
929
0
71
From my recent studies ... It would be fair to say that forcing all cores to be in a nonparked state would yield a tiny bit more performance improvement because the overhead needed to bring them outta parking is of course not there. By overhead, I mean mainly is bringing the cores outta deep C6/C7 C3 sleep states .. which is where they are when parked(if power saving features of CPU enabled). To maintain the full benefit of non parked cores...might as well disable all the cpu power saving features in the BIOS as well =P
If you already have all power savings of CPU disabled in BIOS... parking really gives u no added benefit, so parked / non parked is nearly unmeasurable overhead in that case.

NOTE: I realize sleep states / parking are different mechanisms, one being implimented by hardware and the other by OS for hardware sakes...but they work in relation to each other.

The nice thing about OS method of parking cores though... allows more allotted sleep time for C3/C6/C7 core sleep states to utilizes(more effecient use of the sleep states w/o so many wakeups).
Basically due to how the kernel schedules threads on parked vs nonparked vs 'idling' cores.
 
Last edited:

snoylekim

Member
Sep 30, 2012
104
0
0
Interesting .. I have a box running a Xeon 1245-V2 on Win 8 ..never see any parking in resource monitor ( need to check other utilities to verify) .. draws about 20W at idle .. 3770K on the other machine running Win 7 parks .. draws 9W at idle ..wonder if it's the Xeon,Win 8 or both ??
 

nub1

Member
Nov 16, 2013
33
0
0
I've just found out about this unparking cores things and decided to give it a try.

Here is a graph from Metro LL, as you can see the fps is considerably much more consistent with the cores unparked, the minimum fps has increased quite a bit also.

1569yxw.png


In other games it seemed to have boosted the minimum fps up as-well, no fps graphs like in Metro LL though.

CPU: AMD FX 8350
OS: Win 7 / 64 bit
 
Last edited:

BrightCandle

Diamond Member
Mar 15, 2007
4,762
0
76
We have quite a few studies that have been done with various software that shows core parking on Windows 7 is a bit of an issue. Winrar was where the industry initially saw a problem. Performance was pretty significantly impacted by having the cores park as the algorithm had a tendency to use all cores, then back off and then use them again and that had the cores parking and unparking frequently and the time it takes is not insignificant. The impact was about 30%, which is quite a bit.

So then we had a number of people looking into other problems that it might cause and it turned out a lot of games also had the problem with HT based processors. They would also use multiple cores in periodic ways and caused them to change from parked to unparked frequently and this seemed to impact on the minimum frame rate and the amount of frame time variance the games showed. Some games seemed to gain quite a lot from turning core parking off, BF3 I seem to recall was one of them.

We also know that the various power saving systems in Windows cost performance more generally. Its not just core parking but also the C states. It shows up in benchmarks that there is about 5% difference between balanced and performance power states on a computer in quite a lot of benchmarks, a large part of which can be attributed to core parking but not all of it.

Its no myth, there is definitely a performance impact to core parking and in certain circumstances it can be substantial. So far as I know the problem was resolved with Windows 8 along with a variety of tweaks they did to multithreading in general which make games like BF4 perform quite a bit better on it than Windows 7. Winrar since changed its software as did many of the other benchmarks that were impacted so its actually kind of hard to test it nowadays. But of course its easy to test this yourself in games, just take a frame time trace of a sequence where you think there is or isn't a difference and compare a few runs. Should be easy enough to see if it impacts a particular thing but much harder to see if its fixed more generally.
 

G73S

Senior member
Mar 14, 2012
635
0
0
We have quite a few studies that have been done with various software that shows core parking on Windows 7 is a bit of an issue. Winrar was where the industry initially saw a problem. Performance was pretty significantly impacted by having the cores park as the algorithm had a tendency to use all cores, then back off and then use them again and that had the cores parking and unparking frequently and the time it takes is not insignificant. The impact was about 30%, which is quite a bit.

So then we had a number of people looking into other problems that it might cause and it turned out a lot of games also had the problem with HT based processors. They would also use multiple cores in periodic ways and caused them to change from parked to unparked frequently and this seemed to impact on the minimum frame rate and the amount of frame time variance the games showed. Some games seemed to gain quite a lot from turning core parking off, BF3 I seem to recall was one of them.

We also know that the various power saving systems in Windows cost performance more generally. Its not just core parking but also the C states. It shows up in benchmarks that there is about 5% difference between balanced and performance power states on a computer in quite a lot of benchmarks, a large part of which can be attributed to core parking but not all of it.

Its no myth, there is definitely a performance impact to core parking and in certain circumstances it can be substantial. So far as I know the problem was resolved with Windows 8 along with a variety of tweaks they did to multithreading in general which make games like BF4 perform quite a bit better on it than Windows 7. Winrar since changed its software as did many of the other benchmarks that were impacted so its actually kind of hard to test it nowadays. But of course its easy to test this yourself in games, just take a frame time trace of a sequence where you think there is or isn't a difference and compare a few runs. Should be easy enough to see if it impacts a particular thing but much harder to see if its fixed more generally.
this is by far the most educated answer I've read

thank you