Core i7 Affinity

[Freddie]

I never said they were waiting to be executed...

I mentioned threads waiting to be executed not because you said it, but because I was giving you an example of something that the scheduler is concerned with that does not involve reacting when a core gets close to 100% saturation.

...it just makes it seem like it will fill core 2, then offload to the rest.

I understood that it seemed that way to you. I was trying to explain that in fact, as a general rule, the thread scheduler doesn't do that. And I was also trying to give you an example of why it doesn't do that.

[Freddie]

Obviously there's no science in this...

Very interesting, thanks. I don't know how deeply you want to get into this, but in addition to Task Manager, Windows includes a much more sophisticated program for this kind of analysis. It's called PerfMon. If you use PerfMon you *can* make this science.

If you want me to run any other affinity combinations let me know.

How about graphs for every WoW on all eight cores?

What would also be nice to see -- but this would involve more work -- would be an indication of performance to go along with the affinity setting. You (or somebody) could use Fraps or PerfMon for this.

If you want to get into this more deeply, here's a document about performance tuning from Microsoft that gives some ideas:

http://www.microsoft.com/technet/pro....mspx?mfr=true

I had thought I would need to uprgade from 6GB of triple channel to 12GB based on my experiences with a Core 2 Duo system with 4GB but apparently not (unless I want to turn off my paging file, which I may do in the future).

I'm skeptical that turning off paging can ever be a good idea. The first things I'd do to reduce paging is turn off services and get more RAM.

The document I just linked has a section about memory.

[Creazil]

EDIT: The screenshots were taken at the same location in wow, without moving or anything - In Dalaran with ALOT of people around, if it matters anything.

I made some tests aswell, one with all 8 as you suggested. I had focus on the main, which then gets a max fps of 40 and the slaves 15 fps. The main would be in "01" in the second screenshot and "04" in the third. I didn't bother with "no affinity", cause that just sucks :-)

8 cores:

Second core and last core seems to work a little more than the others.

01234567:

The "main" cores is working more, but the second of the 2 cores is working craazy!

04152637:

Same as 2 pretty much!

The first solution works best, imo :-)

[Creazil]

Compared to the other screenshots, by turbonapkin, it seems like I use alot more memory :-)

[Freddie]

I made some tests aswell, one with all 8 as you suggested.

Thanks. The thing that jumps out at me about that screenshot is how even the load is. Would you agree that it's the most even screenshot so far? It's also the only screenshot where the operating system is completely free to schedule WoW's threads however it wants. In all the other screenshots, the operating system is artifically constrained by the affinity setting. I think this shows what I was trying to say to Clanked, that the thread scheduler tends to distribute loads pretty evenly.

I just want to make sure I understand your other numbers. When you say 01234567, do you mean WoW1 is on 0 and 1, WoW2 is on 2 and 3, etc.?

The first solution works best, imo :-)

This will be very helpful for people to know.

[Freddie]

Could one of you check something? Launch four WoWs from scratch without setting the affinity manually in any way (including not in WoW's config file). Then open the operating system's Set Affinity box to see how WoW sets its own affinities by default. From what we've seen it seems like they will all be set to 0 and 1 but it would be good to know for sure.

[turbonapkin]

Yes, all instances are set to 0 and 1 if you just launch the clients in a normal fashion.

I've been playing with perfmon and a log analyser called PAL, once I have found a way of preesnting the results in a way that may be of some use here I will post back or maybe make a new thread as I don't want to divert from the OP.

[Freddie]

Compared to the other screenshots, by turbonapkin, it seems like I use alot more memory :-)

I just looked for a few minutes to try to find an easy way to judge whether WoW is starved for memory. (I'm looking on Vista.) If you click "Resource Monitor" at the bottom of Task Manager's Performance tab, then expand the memory bar on Resource Monitor, you can see hard page faults per minute for each WoW. That's the number of times the application had to stop what it's doing and retrieve a piece of info from disk because it had been paged out of memory. Ideally that number would always be zero. You can measure the same thing over a longer period of time with PerfMon.

[Creazil]

I just want to make sure I understand your other numbers. When you say 01234567, do you mean WoW1 is on 0 and 1, WoW2 is on 2 and 3, etc.?

Yes.

Yes, all instances are set to 0 and 1 if you just launch the clients in a normal fashion.

That's the same for me :-)

Could one of you check something? Launch four WoWs from scratch without setting the affinity manually in any way (including not in WoW's config file).

How do you do that? - the thing in bold! :-)

As for my performance in general. I am running around with 40 fps in Dalaran on the main (15 fps on slaves), full view distance, full spell details, and full environmental details!

[Creazil]

Compared to the other screenshots, by turbonapkin, it seems like I use alot more memory :-)

I just looked for a few minutes to try to find an easy way to judge whether WoW is starved for memory. (I'm looking on Vista.) If you click "Resource Monitor" at the bottom of Task Manager's Performance tab, then expand the memory bar on Resource Monitor, you can see hard page faults per minute for each WoW. That's the number of times the application had to stop what it's doing and retrieve a piece of info from disk because it had been paged out of memory. Ideally that number would always be zero. You can measure the same thing over a longer period of time with PerfMon.

Ah, neat - it's zero so no problem :-) I've been looking a little at perfmon, i'll check it out tomorrow!