Interesting CPU affinity results from WoW

[jimhsu]

This should go in storage, but load times in WoW are also extremely IO dependent. I can't prove it definitely, but WoW seems to use massively concurrent random reads when doing area transitions, loading, or anytime it needs data. In fact, it's one of the games that benefits the most from a SSD (most modern games don't).

 

[ShawnD1]

I re-activated my subscription to WoW and now I'm looking at this cores thing. I see that this game is still not multithreaded like Blizz says it is. On a dual core E6600, the CPU is pinned to 50% and it sometimes jumps up to 53%. Setting the CPU affinity to use 8 cores probably won't do anything.

 

[konakona]

Originally posted by: Dravic
PH II 955 @ 3.8
4890 @ 950/1100

I actually get better performance when only selecting 3 cores (Affinity = 10) rather then distributing the load to all 4 cores (Affinity = 15). My guess is windows threading over the 4 cores begins to introduce some kind of thread thrashing over using 3 cores.

Beginning Ulduar 25 man before selecting vehicles
Settings everything maxed

Affinity 10 = 71 fps (allowed me to put on 4x AA 8x AF and go down too 51fps)
Affinity 15 = 51 fps

similar results in Dalaran

By any chance, could that have been the amount of L3 cache available per core? I am not sure how exactly L3 cache is managed on PhIIs, so I am really going out on a limb by saying this. Just something I heard when people were discussing the merits/drawbacks of unlocking the tricore 710/720Bs.

 

[Denithor]

Originally posted by: Zap

Originally posted by: Shaq
Nobody posted these yet? CPU benchmarks for WOTLK.

http://www.pcgameshardware.com...-and-Core-i5/Practice/

So, their "SMT" is HyperThreading? Disabling it gains 25% performance?

I'm going to take a guess here and say they probably didn't change the CPU affinity from the default "3" - meaning that with HT enabled they were forcing the game to run on a single physical core (forces the game to run on the first two cores - in this case two virtual cores which are simply the first physical core).

Which would explain why - when they disabled HT - the performance increased by 25%. Turning off HT and leaving the default CPU affinity would result in the game running on two physical cores instead of the first two virtual cores.

EDIT: Confirmed by looking over the rest of the results. The e8400 & q9650 run nearly identical numbers which makes sense if the game is only running on the first two cores of the quad (as two cores of a q9650 is exactly equal to a single e8400).