WinFix: i7 owners Win7 by default is crippling our CPU's

[TheRyuu]

Turning this off would be a good way for a stress test of a selected vcore, making sure that your comp doesn't crash under less than ideal situations - better put that core parking may mask insufficient vcore at certain clocks. That said, its better to leave this on - SMT scheduler is the Linux equivalent, correct?

I'm actually not sure if there is a Linux equivalent, google wasn't very helpful on the matter.

What I think the option your referring to is the SMT awareness of the scheduler such that it prefers a 'real' core if it has a choice (windows does the same thing too).

 

[ShintaiDK]

On Windows 7, it priotizes using physical cores over logical cores because consumer applications tend not to be as well threaded.

On server Windows it turns whatever logical cores off.

Windows Server certainly dont turn logical cores of. And all cores are actually logical. Windows however knows what 2 logical cores share the same physical core. And avoids utilizing both logical cores on that until the other physical cores got 1 logical core loaded each.

It got nothing to do with threading of applications.

 

[magomago]

the 'proof' supporting the link by the original author is amazing!

*that as just a hint of sarcasm*


LOL is all I have to say

 

[IntelUser2000]

Windows Server certainly dont turn logical cores of. And all cores are actually logical. Windows however knows what 2 logical cores share the same physical core. And avoids utilizing both logical cores on that until the other physical cores got 1 logical core loaded each.

It got nothing to do with threading of applications.

There's a difference between Windows Server 2008 R2 and Windows 7.

Windows Server 2008 R2 actually has a feature called Core parking as explained here: http://blog.tune-up.com/news/pdc09-la-day-1-coverage/

What it does is if the current cores that are active aren't fully loaded, it will try to move the workload to those cores and keep the others inactive. And it didn't differentiate between Hyperthreading and physical cores. It wasn't so critical since vast majority of server applications benefited from Hyperthreading significantly and that would in some cases would prevent from actual cores from turning on(because it uses a Hyperthreding one instead).

But in consumer Windows 7, it doesn't do that. The logic was that in cases that the application favors a Hyperthreading core over a physical one there could be performance losses, so rather than doing that the OS priotizes physical cores over Hyperthreading one. It turns on all available cores before using Hyperthreading.

And you can simply observe that in idle. Go to Task Manager and see which logical cores are on and which are off. For my 2600K, it says:

Core 0, 2, 4, 6: On
Core 1, 3, 5, 7: Off

The latter 4 is always off unless I do something that uses more than 4 threads.

 

[garikfox]

Thats not true if you notice it sometimes parks other cores aswell.

Example: When I had core parking enabled I had Core 1,3,6,7 parked, Then after a restart it parked 1,3,5,7

It seems to park different cores not just the HT cores.

 

[IntelUser2000]

Thats not true if you notice it sometimes parks other cores aswell.

Example: When I had core parking enabled I had Core 1,3,6,7 parked, Then after a restart it parked 1,3,5,7

It seems to park different cores not just the HT cores.

That's still only a physical core. When there's only 1 thread per physical core, then Hyperthreading is not being used. Simple as that.

 

[Hatisherrif]

"WinFix: i7 owners Win7 by default is crippling our CPU's"



Is this also the fix for the thing that seems to have crippled your grammar?

First you should spend money on education, i7s will come naturally later

JK

 

[Ryun]

I have a hypothesis about why some people are seeing performance games while others aren't. If you read this page at tomshardware, core parking is supposed to work in conjunction with ideal core: http://www.tomshardware.com/reviews/intel-core-i5,2410-8.html

Ideal core works to keep threads on one core (assuming it's got the processing resources for it) without bouncing it around to other cores. This is similar to the problem the original Phenom I + Vista had where the windows scheduler would bounce threads around and the CPU was constantly throttling its cores speed up and down (resulting in reduced performance).

It's possible that those who are turning off core parking are noticing a gain because ideal core isn't working properly or is otherwise disabled. In my mind what would happen is that windows would bounce threads around across the cores while also parking and unparking them resulting in reduced performance. Subsequently, turning off core parking would alleviate this issue as while the threads would be bouncing around cores, you wouldn't need to wake them up.

Unfortunately, I don't know of where the settings for Ideal core are located or how you would change them so I imagine if I'm correct the only way to potentially fix it would be to reinstall windows 7.

On the other hand, for those who are seeing reduced performance: Monitor task manager with core parking enabled when you notice performance degradation in games. See if you notice threads being bounced around the cores.

Knowing what core parking does, I can't imagine it would result in such a reduction in performance by itself, especially for a multithreaded application like the WinRAR benchmark.

 

[soccerballtux]

I'll try it after Tweakboy tests it out.

agree we need his input on this.