Benchmark

[Jetboy55]

Hi again. I recently tried my hand at benchmarking for first time. What do you all think of 3DMark2000 from MadOnion? It seems like such a straight forward package, and easy to use after reading the manual. I ran the benchmark at 1024x768 16b color, 16b zbuffer, 16b textures, optimized for 3dNow(I have k6-2 processor) and came up with a fine score of 574. Wow, thats bad, right?
I also got my hands on PowerStrip which seems to work well, and I started overclocking my video. From 75mhz engine up to 95mhz, and from 99mhz memory up to 120mhz. Any highter and things get buggy. Ran benchmark again and got score of 660 3d marks, 8.5fps medium helicoptor, and 10.5fps medium adventure. BAD BAD BAD.
What do you all think of this benchmark program, and this utility Im using to overclock? Thanks for your opinion.

[Tiretool]

This is probably the most used universal 3d benchmarking utility used bar none. It is very good in comaring your system ot other systems. The only drawback apparent in this program is that it is very NVidia biased. "Usually" a system using an NVidia graphics card against an identical system using any other card will net a 30-50% increase in the final score. I'm not sure if this is because of fault in the program or if the program runs better on NVidia cards, or what. Other than that... it is top knotch in my book.

With my Celeron 600 overclocked to 982, 128 megs PC 133 Ram, and Elsa Gladiac I get around 4900 3D Marks in 3D Mark 2000, but don't compare your system to this one. Your system and mine are built on two different architectures and are going to spit out different results. Compare your system to other similarly built systems running the same speed CPU with similar video cards and you'll get a better feel of how you compare and how your video overclocking affects your frame rates and your 3D Mark scores.

[Jetboy55]

Yeah, I was comparing my system to similar systems using the online results browser in 3DMark2000. It works just FANTASTIC. I was able to compare also to very similar systems using other PCI video cards, and see that indeed I should see alot of performance increase going from my built in AGP Rage Pro card, to a PCI Voodoo 3 3000. I was hoping this benchmark was a good one, cause its my first experience with benchmarking of any kind. Any other suggestions? How about PowerStrip? I used it to overclock the card. It seems to work pretty good, and specifically supports my video card, and driver. But I dont think any amount of overclocking is gonna make this Rage Pro work well with new games. 9 fps and 11 fps are still both very bad and make game unplayable. Thanks.

[Xayd]

Powerstrip is a great program. I've been using it for quite awhile, and have never had problem one with it. As far as vid card tweaking goes, it's the best there is.

Sadly Madonion has become the major bias-spewing lackey for SSE and Nvidia based systems. For instance, a Geforce 256 DDR in my machine gets around 6500 in 3dmark2000, with a Duron @ 1008 and 128 megs Crucial 2/2/2 PC133 RAM.

We tested the same vid card in a co-worker's P3 733, all other hardware equal, and got over 7000, although my Duron rig blows his away in frames per second in pretty much every game we tried, with the same video settings, same 98FE, and same resolutions.

Xayd

[Jetboy55]

Thanks, keep all this good info coming. Im trying to catch up, just started getting into this kind of stuff.
Can anyone give a quick overview of the major catagories video cards are ranked in? Like fps(I understand this one), triangles, texels/s, pixels/s, and which are most appropiate to pay attention to? THanks a million.

[madthumbs]

Is 3dmark2000 really Nvidia biased, or does it just show the shortcomings of current 3dfx cards? How do ATI, and other cards do on it?

[Xayd]

Well, take a look at the Voodoo5 and the Geforce2 32 meg cards. This has become a "driver race" between the two cards, so lets take a step back and compare them with the original drivers the Voodoo5 shipped with and the pre-detonator3 nVIDIA drivers.

In low detail, the Geforce2 cards got 140-160 FPS in Quake3. The Voodoo5 produced 120 frames per second. At 1024 x 768 with FSAA and all game details on high, the difference in the cards was about 5 frames per second, at 1280 x 1024 and above the difference was less than two frames per second.

So, we have the nVIDIA card winning low res "speed tweaked" tests by 20-40 FPS, all high detail tests ending up virtually equal. There are two low detail tests in the default 3dmark2000 benchmark, which the nVIDIA card should win easily. There are two more medium detail tests that the nVIDIA card should win slightly. This is all rough math, I don't know the specifics, but with the nVIDIA card winning an average of three out of 20 tests, it should have a score roughly 15% higher than the Voodoo5.

I searched Madonion's page for a system with a P3 733, Gefore2 GTS/Pro chipset, Voodoo5 chipset, and default benchmark results. This is what I get...

nVIDIA G2 GTS highest score---6743
Voodoo5 5500 highest score---4301

So, are we to believe that the nVIDIA card is over 37% faster than the Voodoo5? Not in the least. Raw speed indicates that the two cards (not latest drivers, most overclocked, etc., we're talking first generation drivers on each, just the CARD itself) perform equally with all of their eye candy on (both at unplayable framerates with that stuff on, by the way), and both past what the human eye can perceive with all the eye candy off.

And the difference in the 64 meg Geforce2 and the 32 meg Geforce2 in the reviews I saw around was about 5 frames per second in low detail tests and 7-10 frames per second in high detail tests (which is STILL unplayably slow, BTW). So the "yeah but 64 megs" argument doesn't hold water, especially since the Voodoo5's and the 32 meg Geforce2's cost the same amount of money.

Bottom line is the two compare comparably. The Geforce2 Ultra is much faster I'm sure, but in the 200-300 dollar range, the difference is nil. With the Geforce2, you get Hardware T&L, now all you need is a game that utilizes it. With the Voodoo5 you get Glide, good luck on finding any new games that utilize it.

I consider it a toss up, unless 3dmark2000 is your best "game" .

Xayd

[Jetboy55]

Is fps the only perfromance criteria that people who "know" pay attention to? Could you explain what pixels/s, texels/s, triangles/s, polygons/s, mean, which is most important to look at, what a texel is? Thanks.

[Tiretool]

This may help explain some of your questions. It deals with the Geforce 2 directly but I think it will give you a clue about some of the other questions you've been asking.

Hope it helps! If you're still in the dark then say so and we'll break it down a little more.

[Jetboy55]

That helped alot for me to understand what a video card is doing, and what different components of video card are for. But is a texel just a pixel with a texture put on it? If a quad piped engine can render 1 pixel per pipe per clock cycle, then times that by the speed of the engine, you get the theoretical MPixels/s of the card, right? Then if we know that the card can also apply 2 textures to each of those pixels at the same time, we could assume that the MTexels/s would just be double the MPixels/s, right?(Theoreticaly) Am I following this correctly? Please advise if Im on the right track. Thanks again.