Analysis of affinity

[KimB]

In response to the discussion that occurred in this thread, I did a little bit of analysis with my currently-installed software on how well tasks stick to one core.

Here are my notes:

BG: One CPU
BG2: One CPU
Doom3: Fair use of dual-core (90%-50%)
FEAR: Not much CPU usage: none of second core
FF7: One CPU
GTA:SA: Mild switching (80%-20%)
Morrowind: Mild Switching (80%-20%)
NWN: Mild Switching (80%-20%)
GalCiv: Altarian Prophecy: Little Switching (90%-10%)
zSNES: One CPU

Some things to note: all analysis is subjective, making use of the Windows task manager for a few seconds of gameplay while no other processor-intensive programs were running. When I state "one CPU," then usage of the second CPU is only a couple percent. While it might be interesting to go back and look at the performance of games where there was some apparent task switching, the three where it may have been of issue aren't easy to benchmark (and neither is performance much of a concern, fortunately).

Please note that in all cases where it looked like there was some task switching between the CPU's, one CPU seemed to remain dominant.

So, from this little analysis, it seems to me that there isn't a significant problem on Windows XP SP2 of tasks switching unecessarily between processors.

 

[Rys]

Athlon 64 X2, right? Quick question if you do: do you install the X2 processor driver for XP, and any CnQ driver or software, or set CnQ in the BIOS at all? Interesting results since I've been working on something related in recent weeks, cheers for sharing.

 

[KimB]

Athlon 64 X2, right? Quick question if you do: do you install the X2 processor driver for XP, and any CnQ driver or software, or set CnQ in the BIOS at all? Interesting results since I've been working on something related in recent weeks, cheers for sharing.

You know what? I don't remember whether I did install the X2 processor driver or not (rather, I know I did at one point, but I think I've reinstalled Windows since). I did, however, install the performance fix for single-threaded apps on multicore processors. I haven't touched Cool 'n Quiet at all.

 

[Gubbi]

I did, however, install the performance fix for single-threaded apps on multicore processors.

This might be why I'm seing different results. Mine is always split 50/50%. I just dropped in a different CPU and assumed it would work, bit naive I guess.

Cheers

 

[KimB]

This might be why I'm seing different results. Mine is always split 50/50%. I just dropped in a different CPU and assumed it would work, bit naive I guess.

Well, it shouldn't be different because of this. The fix was described on Microsoft's site as basically being related to different clock-throttling behavior.

 

[Gubbi]

Well, it shouldn't be different because of this. The fix was described on Microsoft's site as basically being related to different clock-throttling behavior.

Bugger. I'm not running any throttling software (CnQ) at all, so couldn't be that.

Back to square one I guess.

Cheers

 

[KimB]

Well, XP with SP2 has clock throttling behavior enabled by default.

 

[Diplo]

You guys should check out this article on Tom's - it includes a handy page about a utility that can permanently allocate affinity of a CPU/Core.