Until today I had fooled around with overclocking only to see how hard my video
card could be pushed before it went unstable. That is, I overclocked only to
get a "feel" for the level of quality of my hardware and drivers.
My system includes some relatively high-end components, and since it has always
previously been "fast enough", I haven't bothered with overclocking for sake
of performance.
Of late, however, I've been playing Tribes
2 (T2) quite often, and as anyone who plays the game knows, it's a bit
of a pig. Even on my relatively high-end system, I average around 30 - 45 FPS
in most situations with all of the detail settings cranked to the max. That's
entirely acceptable as far as I'm concerned, but I've recently discovered that
it bogs down unacceptably on certain maps and in certain situations. I could
always turn down the detail levels, of course, but the game is utterly gorgeous
with all the settings maxed out, and so doing would obviate my reasons for buying
"big iron" in the first place.
Until yesterday I was using the latest official drivers (viz., v6.50) from Creative Labs for my Annihilator Pro 2 video card. I decided to install the latest reference drivers from NVIDIA (viz., v12.41) instead, however, and tweak the system to see just how much more performance I could squeeze out of it. My ultimate goal, really, was to render T2 playable under all circumstances in which I was previously seeing a slowdown, but I'm writing up my findings because I made some worthwhile discoveries along the way.
Before I present the data, some discussion of how it was obtained with the benchmarks I'm using seems appropriate. Because my concerns about system performance deal almost exclusively with games, I am using three different benchmarks for my testing: (1) 3DMark2001 by MadOnion, (2) Quake III Arena Demo (Q3A) and (3) Unreal Tournament (UT) v4.36. When taking benchmark data, I always follow the same procedure:
By always following the same procedure, I hopefully minimize any artifacts from my operating system and at least achieve some degree of consistency for sake of comparison. It also bears mentioning that I run both the Q3A demo and the UT demo at 1024 x 768 x 32 bpp resolution with all of the graphical options set to maximum. This is the resolution at which I typically play, so it (hopefully) provides a genuine reflection of the kind of performance I can expect at game time. For sake of reference, I think it useful to provide a more complete picture of my system, which includes the following components:
My system includes other components as well (e.g., a Wacom graphics tablet, a Microsoft Sidewinder Strategic Commander, etc.), but only the above components seem more relevant to my testing as only their drivers and so forth are loaded while benchmarking. Finally, I am using the latest version (viz., v4.31a) of the VIA 4-in-1 driver pack for my motherboard chipset under Windows 98 with all current patches for that operating system.
The following chart is taken from an Excel spreadsheet of mine, which documents the various changes I've made and the various changes in performance I've measured. In all honesty, the UT benchmarks are probably not all that useful as UT is far more CPU-bound than video-bound. Still, I may yet overclock my CPU in the future, so perhaps having the data will eventually pay off.
| Date | GPU/Mem. | 3D Marks | Q3A | UT | Comments |
| 3/31/01 | 200/333 | 2844 | 92.5 | 40.20 | Switched the motherboard to an ASUS A7V133 |
| 3/31/01 | 200/333 | 3053 | 99.1 | 40.40 | Enabled DMA access for the hard drive |
| 5/27/01 | 200/333 | 2462 | 99.4 | 40.69 | Updated to VIA 4-in-1 drivers v4.31a |
| 6/15/01 | 200/333 | 3243 | 99.1 | 40.11 | Switched to NVIDIA v12.41 reference drivers |
| 6/16/01 | 210/370 | 3429 | 107.2 | 39.71 | Overclocked the video card for better performance |
In the above chart, the date reflects when the data were taken, the "GPU/Mem." column shows the clock settings for the video card's core (i.e., Graphics Processing Unit) as well as its memory, and the remaining columns should be pretty self-explanatory. It is worth noting that my system could actually complete the benchmarks with clock settings as high as 210/380, but I observed some undesirable visual artifacts at those speeds and thus had concerns about stability. The extra 10 MHz. of memory-clock speed isn't worth risking the video card in my estimation.
Note well the difference it makes to use DMA access for the system's drives (roughly 7%), as well as the surprising drop in 3D Marks when updating the VIA drivers from the previous version I had in place (v4.28). I was quite surprised to find that the latest release of the NVIDIA reference drivers gave significantly better performance with the 3DMark2001 benchmark, more than restoring what was lost in updating the VIA chipset drivers. The final results are still more impressive. After switching drivers and overclocking the video card, I achieved an overall boost of almost 40% with 3DMark2001 and almost 8% with Q3A. I am somewhat at a loss to explain the slight drop in measured performance with UT, but as I said previously it doesn't seem particularly relevant. That is, all of the UT scores are so close as to make the differences statistically insignificant.
Still more impressive, though admittedly more difficult to quantify, are the changes I've seen with T2 and UT. At the present time, I confess that I neither have any benchmarking demo for T2 nor would I know how to use it if I did. I am a relative newbie to the T2 community and as such have no idea how to take benchmarking data with it. Still, I can report some general observations. First, I no longer experience unacceptable slowdowns when playing T2 with the previously problematic maps and situations. The game seems to run more smoothly overall. Whether this is due to the change in drivers or the overclocking I cannot say as I did not take any T2 benchmarks or even play the game between the two changes. Rather, I fired up T2 after both the driver switch and the overclocking and played several games, and not once did my frame rate drop unacceptably. Second, if observations of the average frame rate are at all worthwhile, whereas I used to see between 30 - 45 FPS most of the time, I now see between 50 - 70 FPS most of the time. This isn't "hard data" by any stretch of the imagination, but it seems unavoidable that the driver switch and overclocking have given T2 a "shot in the arm" with my present system.
Further, since the changes I've noticed smoother performance in UT as well. Previously my system would bog down at times when in large, open areas with a fair amount of geometry and lighting. For example, when standing on the sniper deck in the central area of DM-Liandri, my frame rate would sometimes drop noticeably such that gameplay would become jerky—particularly if many players were present. But though the UT benchmarks indicate a small, negative effect, experience indicates a positive change. To be more specific, I've actually increased the resolution at which I run UT to 1280 x 960 x 32 bpp with all graphics settings maxed out, and I'm not seeing any unacceptable slowdowns. I am finding this somewhat difficult to believe, but I've played several games of UT on-line after making the changes, and in each case I haven't seen any problems—even on games with as many as sixteen players visible at once. I cannot explain the differences, but I sure don't mind (grin).
While I haven't overclocked my hardware in the past, and while I have tried to stick with the manufacturer's recommended drivers, I am going to go forward using the reference drivers for my video chipset and overclocking insofar as my system remains stable. I will likely try overclocking my CPU as well in the not-so-distant future. I have no idea whether T2 is CPU-bound or video-bound, but you can bet I'm going to try to find a way to make that determination. And if it's CPU-bound, I just might end up pushing my Athlon chip a bit. After all, if T2 is gorgeous at 1024 x 768 x 32 bpp, how might it look at even higher resolutions? (grin) I know, I'm spoiled. It's a good time to be a gamer.
06/16/2001