Windows 7 SP1 performance increase for CQD owners

I will not take credit for the find so I am quoting from the original thread in gateway forums.

It is not the full windows 8 scheduler but for single threads it does give a 5% performance increase but nothing for heavilly multi thread. If AMD gets M$ to fully release the new Windows 8 scheduler for Windows 7 expect 10% increase for any app that does not already optimize multithread by internally setting core affinity. IE 3DMark06 CPU score at 3.2 GHz here goes from 4500 on windows 7 to 4950 on Windows 8 task scheduler.....................

Edit; I have put this here as this is specific to CQD owners, this will not help the older C2D owners as far as I can tell...............

 

Anandtech benchmarked these fixes a while back:

AnandTech - The AMD FX (Bulldozer) Scheduling Hotfixes Tested

I don't get how this is relevant to laptop owners or NBR in general though. AFAIK, there's no Bulldozer laptops and Trinity won't be coming out for at least a few months. Is the CQD a new Gateway laptop sporting a desktop Bulldozer CPU?

 

Right for bulldozer. CQD is much older tech, the quad core CPU's. Like mine the q9200 but that is ES, there were the official q9000, q9100 and qx9300. These older CPU's are also SMT but unlike the iCore that supports NUMA. The windows 8 scheduler supports enhancements for pre NUMA multicore CPU's and the bulldozer enhancement patch is just some of the enhancement.

Edit; for the full enhanced Sheduler to function you would need to loose setting affinity as if affinity is not programatically set it would then confuse the scheduler. if the program passes the affinity and lockout there probably is not an issue. if you force it despite the scheduler there would be latency issues. This is probably why with Windows 8, at least with my installs, there was no affinity option in task manager.

 

Can you go into more detail about what you mean, and what should be done on a Core 2 Quad after you install the update?

 

Just install the patch. It will give about 5% better times in single thread such as SuperPI if you need to see it. Now this present patch does nothing for heavilly multithreaded apps such as 3DMark06. That type of program is heavily threaded but relies on the OS scheduler rather than the program selecting affinity for its threads.

Even with Windows 8 scheduler where affinity is set for a thread programatically there does not seeem to be a performance gain. This was a very short impression and I'd have to test that out further as my only program I use is Canon DPP but that has alot of I/O.

Yeah I though of using C2Q but it really is a C2D2 as there really is on the chip 2 seperate C2D's. Sorry for the confusion.

Since Windows 7 is NUMA aware it will not even with the PRO version alow you to split the single chip into multiple groups. If this could have been done then the scheduler could optimize for the CPU.

What happens behind the scenes is when a thread splits between core 0, 1 and then core 2 and/or 3 then everything has to load up in the second cores internal memory. This cause a huge latency issue for threads that are not affinity set programatically.

Now again this patch starts to address the issues but not fully. Windows 8 scheduler does fully address the issue from what I can see. If you own one of these older quad cores Windows 8 will provide a nice CPU boost in most cases. I am hoping AMD will push M$ to release it as a patch for Windows 7 and Bulldozer but I am not holding my breath.................

 

I don't think anybody really cares about how Intel put the cores together. I mean, everybody here knows that Arrandale's GPU is on a separate die, but we all say it's integrated into the CPU anyway. At best, these sorts of things are interesting bits of trivia, at worst, it's nitpicking.

Either way, clarity is far more important and sticking to established convention guarantees clarity.

 

True but that is a GPU, seperate functions. Naming conventions aside all I care about is performance or lack there of...............

 

Finally found the thread where I had mentioned the issues and possible solutions 11 months ago................

http://forum.notebookreview.com/gateway-emachines/570826-c2q-not-c2q-question-imho.html