Gaming-Sources r2 Hz Tuning Performance

miseiler · Tux's lil' helper Joined: 17 Mar 2003 Posts: 118

For those unaware, the gaming-sources-2.4.20-r2 ebuild bumps the ck kernel version to ck6. Among other things, this adds a Hz tuning option allowing one to change the number of times the system timer is accessed per second.

The vanilla kernel option default is 100 and the ck (and thus gaming-sources) default is 200. It gives some suggests for personal tuning in the help file, which I'll quote here:

neuron · Advocate Joined: 28 May 2002 Posts: 2371

from a programers point of view I have to ask, why would increasing the number of times the system timer is accessed (unless the number is very low, which would could screw up threading or something like that) increase performance? heh

and I'd try, but this damn nforce2 motherboard has no agpgart until nvidia gets it going

Lovechild · Posted: Sat Apr 19, 2003 7:53 pm Post subject:

The reason we don't want to probe the timer to often in 2.4 is that it causes a horrible overhead - in 2.5 this is much improved.

anyways, in some cases, probing more often can lead to a smoother system, because you get a more finely grained timing system.. at the cost of throughput of course.

MadEgg · l33t Joined: 06 Jun 2002 Posts: 678 Location: Netherlands

I'm running gaming-sources with Hz set to 500. I do notice an performance increase over my last gaming-sources, but that was gaming-sources-r1...
_________________
Pentium 4 Prescott 3,2 GHz
Asus P4P800 SE, i865PE chipset
1024 MB PC3200 RAM
AOpen Aeolus GeForce 6800 Ultra 256 MB DDR2
Creative Audigy2 ZS
gentoo-sources-2.6.20-r7
nVidia-drivers version 9755

miseiler · Tux's lil' helper Joined: 17 Mar 2003 Posts: 118

Ok boys and girls, I have returned with benchmarks!

System:
Athlon 1200 Mhz
GeForce3 w/ nVidia 4349 drivers
1 Gb PC133 RAM
Maxtor 40 Gb ATA100
Desktop color depth 16

Kernels Tested:
Gaming-sources r1 (ck4 patches)
Gaming-sources r2 200 Hz (ck6 patches)
Gaming-sources r2 500 Hz (ck6 patches)

Quake 3 v 1.32b was used, with the default demo "four.dm_68".

In all tests, the "High Quality" settings were set, with the resolution being the only independent variable. No AA or AF was used.

Gaming-sources r1
------------------------
640x480 137.6
800x600 137.3
1024x768 133.4
1280x1024 118.4
1600x1200 106.0

Gaming-sources r2 (200 Hz)
------------------------
640x480 137.2
800x600 136.7
1024x768 133.1
1280x1024 118.4
1600x1200 105.6

Gaming-sources r2 (500 Hz)
------------------------
640x480 135.6
800x600 135.0
1024x768 132.5
1280x1024 117.1
1600x1200 105.2

Across the board, there isn't a single instance where we find a statistically significant change in fps. The only thing I noticed when testing kernels is the 500 Hz version *seemed* to compile faster. I didn't do any actual timing on it, though.

Those of you who understand the way this works, is there any reason a higher rate would decrease compile time?
_________________
Cthulhu for president.

zephyr1256 · Posted: Sat Apr 19, 2003 10:56 pm Post subject:

Does this Hz value just indicate how often the CPU does context switches because the timer runs down? If so, I wouldn't expect to see much difference. Perhaps you would see an improvement(in fps, and/or responsiveness) if you try running your process at the same time some other CPU hog(like zetagrid, shameless plug) is running. You're getting less total work done on the CPU(usually not a concern with desktop systems) because the CPU is doing more context switches than at lower Hz.

Migrant Programmer · n00b Joined: 31 Aug 2002 Posts: 10

miseiler · Tux's lil' helper Joined: 17 Mar 2003 Posts: 118

Figures they'd do that.

barlad · l33t Joined: 22 Feb 2003 Posts: 673

I get about 200 fps less in ck6 kernel. Out of 5000. It is not really that big but overall I am more satisfied with ck4 than with ck5 & 6, especially stability-wise.

miseiler · Tux's lil' helper Joined: 17 Mar 2003 Posts: 118