Core Parking

[chazus]

Well, it's that time again when I see posts regarding something and realize... I have little to no clue as to what something is.

Core Parking.

As far as I understand it, it has to do with how Windows manages cores being active, and sometimes gets a bit too zealous, leaving cores disabled when it could be using them more efficiently.

Thats... about as far as I understand it. Modifying that info requires editing power profile settings (?)

My question is this.

I often see Core Parking as the suggested culprit of poor game performance. Given that I've not modified my settings to my system, why am I not experiencing problems, while other people (often with better systems) are? I guess I'm wondering if Core Parking is a sort of red herring issue, or maybe I just don't understand it clearly enough.

[Hottage]

Very few PC games are optimised to make use of more than two cores anyway, so Core Parking is unlikely to be a huge issue.

[Masoner]

Task Manger > CPU > Open Resource Manager > Look @ each CPU and see if it says "parked"

[n0cturnal]

Core parking is normally not a problem on a fairly new computer but sometimes it can go wrong and cause some performance issues.

http://bitsum.com/about_cpu_core_parking.php

[Masoner]

Core parking is normally not a problem on a fairly new computer but sometimes it can go wrong and cause some performance issues.

http://bitsum.com/about_cpu_core_parking.php

that is a nice program, I just ran it and all is flickering green.

[Cyanotical]

I often see Core Parking as the suggested culprit of poor game performance. Given that I've not modified my settings to my system, why am I not experiencing problems, while other people (often with better systems) are? I guess I'm wondering if Core Parking is a sort of red herring issue, or maybe I just don't understand it clearly enough.

people lie, and/or are dumb, my best example are the people claiming their alienware is getting 200 FPS on ultra when my system is crawling along at 45 (timeless isle launch, locked server)

core parking is a hold over, and is not needed on anything since vista, essentially you could lock a process to only use a certain core, this was useful on dual cores so that you could put something like a render on core1 and everything else on core0

its not needed anymore simple because SMP kernels are much more mature and handle hundreds to thousands of threads without a problem, parking a process (and subsequently all its threads and handles) on a single core will in fact reduce performance on modern computers


it still shows up from time to time when people do some digging for ways to improve performance without spending money

[chazus]

but sometimes it can go wrong and cause some performance issues.

it still shows up from time to time when people do some digging for ways to improve performance without spending money

See, that's the issue. I'm talking about people with 3570K+ CPU's having massive FPS issues. How exactly does it 'sometimes go wrong', is it a problem with windows services? Wouldnt setting to default fix the issue (rather than modifying it manually).

I'm just confused as to why something that works in its default state would cause problems and require modification.

[Cyrops]

I want to do some tests with my rig and core parking, can anyone suggest a software for that?

[Cyanotical]

See, that's the issue. I'm talking about people with 3570K+ CPU's having massive FPS issues. How exactly does it 'sometimes go wrong', is it a problem with windows services? Wouldnt setting to default fix the issue (rather than modifying it manually).

I'm just confused as to why something that works in its default state would cause problems and require modification.

most gaming and enthusiast boards default to C6 off in the bios, which eliminates the sleep status of cores, but it doesn't stop people from parking threads on a core, or at least trying to

i think part of the problem is that people simply expect too much from a new computer, ie "wow is 10 years old, you can run ultra on a potato" not realizing the limitations of the way the engine works

another part is the standard stuff, win7, dx11 game, youtube running with flash hardware acceleration on, etc

[Krasuss]

See, that's the issue. I'm talking about people with 3570K+ CPU's having massive FPS issues. How exactly does it 'sometimes go wrong', is it a problem with windows services? Wouldnt setting to default fix the issue (rather than modifying it manually).

I'm just confused as to why something that works in its default state would cause problems and require modification.

iirc the unpack app modifies registers, fixing issue when sometimes all cores doesnt wake up when needed causing bad performance. I have no idea how it works exactlly, but considering it is software based i believe it can randomly get messed up like every other drivers or something and cause issues.

I have 3570K without issues, but the race for maximum performance in BF4 in its early unoptimized state made me use it and never look back. I think the placebo effect is strong in this one lol

[Cyrops]

Ok, so I ran tests and I got clear results. When parking was enabled 4 cores were parked all the time (during testing).
Used Shrine to Binan village flightpath as a test.

Single WoW:
From Shrine:
Unprk Avg: 54.842- Min: 30 - Max: 88
Parked Avg: 57.884 - Min: 37 - Max: 87

To Shrine:
Unprk Avg: 51.279 - Min: 31 - Max: 87
Parked Avg: 53.891 - Min: 31 - Max: 93

Dual WoW:
From Shrine:
Unprk Avg: 34.412 - Min: 19 - Max: 56
Parked Avg: 29.453 - Min: 18 - Max: 51
To Shrine:
Unprk Avg: 33.979 - Min: 19 - Max: 56
Parked Avg: 29.004 - Min: 16 - Max: 45

We see clear FPS increase with 4 cores parked on single instance of WoW, but when you run two WoW's, disabling parking increases performance.

My theory behind that: when you have parked cores, single threaded applications get a boost and when you unpark cores, multithreading gets a boost.
In my opinion core parking is nice feature for single threaded applications, but due to some 'bug' or oversight, the cores don't get unparked efficiently. For example if I didn't have 4 cores parked running double WoW I would have perfect system (I know I'm exagerrating, but bare with me).

My conclusion: if you have a fast CPU it's more efficient to let the thread 'wait for it's turn' than 'waste resources' assigning it to another core. In other words, threads get processed faster without switching load to other cores. But only until certain amount, as you can see from the test.

It's only my speculations though.

I run i7-4771 and I was using integrated GPU, so it's all CPU. It might be more accurate results without GPU or on the contrary it might skew results.

- - - Updated - - -

iirc the unpack app modifies registers, fixing issue when sometimes all cores doesnt wake up when needed causing bad performance. I have no idea how it works exactlly, but considering it is software based i believe it can randomly get messed up like every other drivers or something and cause issues.

You can only turn it on or off, you can't adjust when the cores get parked/unparked.
You can adjust the max/min amount of cores parked (you can simulate less cores on your CPU this way also).

[n0cturnal]

How many times did you run those test? Also you seem to have swapped the numbers or something since single WoW had better performance in unparked mode and dual had better in parked mode.

Most of the difference could probably be summed down to margin of error as well.

[Cyrops]

How many times did you run those test? Also you seem to have swapped the numbers or something since single WoW had better performance in unparked mode and dual had better in parked mode.

Most of the difference could probably be summed down to margin of error as well.

Ugh I'll have a second look, was editing numbers too much (couldn't find a neat way to represent results).
Yea, oops, mixed parked/unparked names. Will fix.
Well, I ran the test once, obviously as I couldn't handle sitting there longer
I know the results aren't super accurate, but we can see the trend.

The difference is obvious. Can you tell me a 'faster' way to benchmark many times? I remember there used to be in Quake games some console command to test FPS (damn I feel old -_- ).

[Gravewyrm]

I recently ran into a core parking issue with my new build. Its an AMD platform, and after I killed parking via a registry edit, surprise! no more frame stuttering in games. I don't claim to know how it exactly works, but for a user with an AMD APU, having both turbo and the IGP disabled, setting the registry settings from default value of 64, to zero, gave me smooth gameplay.
I'll edit this post when I dig up the place I found the pertinent info from.

:::Edit::: Here's where I got the info about registry edits. ----> http://forum.cakewalk.com/Windows-7-...-m1861804.aspx

[glo]

most gaming and enthusiast boards default to C6 off in the bios, which eliminates the sleep status of cores, but it doesn't stop people from parking threads on a core, or at least trying to

i think part of the problem is that people simply expect too much from a new computer, ie "wow is 10 years old, you can run ultra on a potato" not realizing the limitations of the way the engine works

another part is the standard stuff, win7, dx11 game, youtube running with flash hardware acceleration on, etc

Huge performance drops are common with core parking enabled, though. Mostly on hyperthreaded processors. When you're system is essentially shutting down an entire core of an i3, performance plummets. When your i7 suddenly becomes only a dual core, same issue.

[Cyrops]

/timetest was what I was looking for
I'll do the 'honest person' approach and sit there for 15mins *cough*
Once I'm back home, that is.

[Deleted]

/timetest was what I was looking for
I'll do the 'honest person' approach and sit there for 15mins *cough*
Once I'm back home, that is.

Just make sure you follow EVN's way of testing to the point and probably a few extra tests on top of that or he might not approve

I look forward to the results though.

[Cyrops]

Just make sure you follow EVN's way of testing to the point and probably a few extra tests on top of that or he might not approve

I look forward to the results though.

I was using FRAPS, but didn't know the /timetest

[Deleted]

...
core parking is a hold over, and is not needed on anything since vista, essentially you could lock a process to only use a certain core, this was useful on dual cores so that you could put something like a render on core1 and everything else on core0
...

Think you're confusing core parking with dynamic core assignment.

Core parking would be disabling one or more cores if they are no tasks requiring them running.
What you are talking about is dynamic core time allocation or w/e the name is, anyways it's about OS executing task X on core Y then moving it to core Z and back to Y and so on and so forth, why is it done? I'm not sure.

[Hyve]

Very few PC games are optimised to make use of more than two cores anyway, so Core Parking is unlikely to be a huge issue.

This is very true. Very few games ever make full use of more then 2 cores, or the 64-bit technology.