Print this article

Detonator Performance Compared


Date
: 03/13/01 – 09:17:47 PM

Author
:

Category
: Drivers

It seems as though in recent history, NVIDIA has had a bit of a security problem.  New graphics card drivers have somehow unofficially found their way into the hands of many thousands of people.  Now, whether or not this has been encouraged or premeditated by NVIDIA, those people are probably glad they got their new drivers.  With all the releases we keep getting, however, it’s hard to keep track of which is the one to use.  You would think the newer the better, right?  Maybe not always.  You might download and install some brand new ‘leaked’ drivers, only to find that Quake3 runs like Sally Struthers after a turkey dinner.  Well that’s what this article is for.  No, it won’t go over every single release, but it will go over the major ones.  This article will mull over the performance aspect of the latest major ‘leaks’.

Since there isn’t really much explanation needed to get started here, and you’re all here for the same thing, let’s get started.

The drivers we’re going to discuss are the 5.32, 6.62 WHQL, 7.52, and 10.80 detonators.  The <5.x detonators have been left out due to the fact that the Geforce 2 MX(which is the test card here) is not supported by drivers before 5.32(or so I believe).  The test motherboard in this case is an ASUS P3V4X(VIA 133A chipset) running the VIA 4.29 4in1 drivers.  If you’d like to see an earlier VIA AGP driver comparison, click here.

As benchmarks, I used Quake3: Arena to test OpenGL, and 3DMark2000 for the DirectX side of things.  Since I have nothing else to prep you with, here are the Q3A settings I had.

Windows 2000 SP1
CPU	P3 550E @ 825mhz
SOYO 6BA+IV(BX)	ASUS P3V4X 1006T BIOS
RAM	256MB PC133 SDRAM 2-2-2-5 @ 112mhz
V. Card	Innovision Tornado Geforce 2 MX @ 175/166
4in1	4.29
Benchmarks	Quake3 1.17 demo001.dm3 and 3DMark2000

System Settings
GL Driver	Default
GL Extensions	On
Resolution	640×480 to 1280×1024
Color Depth	16/32bit
Fullscreen	On
Lighting	Lightmap
Geometric Detail	High
Texture Detail	Highest
Texture Quality	32bit
Texture Filter	Trilinear
Graphic Options
Simple Items	Off
Marks on Walls	On
Ejecting Brass	On
Dynamic Lighting	On
Identify Target	On

Now that that is all straightened out, here are the Quake3 results.

Take your time and look over the whole thing.  Instead of going through each variation and commenting, I’ll go over the basic patterns I see throughout, and my thoughts on what they mean.  If you’ve got some ideas, don’t hesitate to leave them in the comments.

5.32
– The 5.32 drivers, which have been obscure for almost a year, head up the rear throughout the entire process.  There is a fairly major performance drop going to 32-bit, even at low resolutions.  Not only this, but the amount of control over such things as anti-aliasing is pretty pore.

6.62
– The 6.62 drivers display a strong showing through the entire barrage of tests.  They ruled over the 5.32 results in all the tests, and offer some great options for you to play with in the properties.

7.52
– The 7.x driverset was the first to require(for the most part) DirectX.  Some claimed 3DMark2000 performance enhancement, others just reinstalled their 6.x set.  When the 7.17s came out, and I installed them, strange things happened to my Quake3 installation.  Luckily, the 7.52 set is pretty much perfect.  There is nothing to complain about here, as far as performance goes.  While not offering any gain over 6.x, DirectX 8 support helps for all you Half-Life/UT players out there.  I consider both games pretty much irrelevant to video card benchmarking, so there aren’t any results from those games.

10.80
– I don’t have a huge amount of knowledge about this set.  There wasn’t much news about them that I saw, and didn’t even know they existed until the first 10.x Win2k drivers came available(which are these).  Their performance, as you can see, is certainly strange.  32-bit performance visibly suffers, while 16-bit performance is very strong.  My 1024x768x16 scores when overclocked to 220/200 were above 80fps for the first time ever.  I was definitely pleased at that little tidbit.

It’s time for the 3DMark2000 results.  Please refrain from laughing at my scores here if you’re one of those 8000+ people.  Some of us can’t afford the equipment needed.  I ran the default benchmark, which is at a resolution of 1024x768x16.

The first thing you’ll most likely notice is the lack of 5.32 results.  That’s because I couldn’t get 3DMark to complete the complete run of tests.  It kept rebooting itself after 5 or 6 tests.  I couldn’t get it to run no matter what I did.  The others worked just fine, however.  While the 7.52 drivers, who supposedly boost 3DMark performance, did not boost performance, the 10.80 drivers did to a fair degree.  

There are a lot of reasons why I don’t like running 3DMark as a benchmark, but the main one is that it isn’t real.  So you got 5000 CPUmarks.  What does that mean?  How much better is it to get 5225 CPUmarks?  As far as I’m concerned, it’s just an arbitrary number.  Comparisons help, however.  If you see a guy with 4900 CPUmarks, and you’ve got 5000, you know you did better than him.  Aside from that, though, there isn’t much use for it.

So which driver set should you use?  The most responsible answer to that would be the latest official NVIDIA release, which as of now are the 6.50 drivers.  As far as performance and hard numbers go, if you’re running an SDR Geforce or Geforce 2 MX, I’d say run at 1024x768x16 and use the 10.8 drivers.  That’s what I’m doing, and I couldn’t be happier.  Overclocked to 220/200, and with all the game settings to highest quality, I’m getting above 80fps at that resolution.  This is an arbitrary number for you though, so I suggest going into Quake3(if you have it, that is) and trying out the settings I have listed above.  Then, to run the timedemo benchmark, go into the console(press the tilde[~] button at the top right), then type
timedemo 1
.  Press enter and type
demo demo001
.  After entering again, it’ll go through the demo and go back to the front menu.  Open up the console again and there are your frames per second.  Anyone with a DDR Geforce or better should easily top my score.

For you people who treasure your 32-bit performance, and would also like whatever DirectX8 features the 7.x series gives you, go for that set.  For those of you interested foremost in stability and assurity(is that a word?) that you won’t crash due to your video card drivers, go for either the 6.62 WHQL drivers, or the official 6.50 leaked reference set.

And for you sickies that like lesser performance for the sake of some sort of demented nostalgia, go for the 5.x drivers.  The run slower, and are less stable, so you should be satisfied.

In conclusion, most of the leaked drivers out there don’t offer much more than bug fixes.  If you’re lucky(or unlucky, depending on your POV), you’ll get a set of drivers that fixes any bugs you might have.  Others, even luckier, will get a performance boost because of one thing or another.  The other unlucky people will do a lot of rebooting.

Get my excel spreadsheet here.