3DMark vs Doom3

[Greg]

Now that 3DMark2003 is old and the new 3dMark2005 just around the corner, we can look back and decide if 3DMark correctly informed, or misled consumers.

This will naturally be a contentious topic. In my mind, 'new/current generation' games are those with per pixel lighting and shadows, no matter what the method is used, such as bump mapping for lighting and stencil or texture shadows. I'm sure others will argue that new games would have to use floating point color/texture formats, or other features, perhaps not available in older video cards.

We could compare a few games like FarCry, Doom3, DeusEx, Theif 3, Battlefield Vietnam with the results from 3DMark game tests.

This test is 3DMark Battle of Proxycon vs Doom3:

System 1
GeforceFX 5800 (@485/970)
Athlon64 3200+
1gb PC3200

3DMark2003 GT2 800x600: 48.7 fps
3DMark2003 GT2 640x480: 58.9 fps
3DMark2003 Default score: 5641

Doom3 800x600xMed: 68.2 fps
Doom3 640x480xMed: 81.2 fps

System 2
Geforce3 ti200 (@220/480)
Athlon XP2100+
512mb PC2100

3DMark2003 GT2 800x600: 11.9 fps (Expect ~17.3fps)
3DMark2003 GT2 640x480: 14.0 fps (Expect ~22.8fps)
3DMark2003 Default score: 1297 (Expect ~1975)

Doom3 800x600xMed: 24.3 fps
Doom3 640x480xMed: 31.0 fps


I would say the 3DMark frame rate score and the overall score which is all most people would see is a little misleading but not that bad. Unfortunately I don't have the Duron machine up to compare video and cpu mixes. What is interesting is that a fast CPU combined with a old video card allows current games be be played quite well, though at reduced resolution. Since the rendering cost has increased per pixel, this makes sense. This is not obvious from the 3DMark scores.

The method I used to compare scores was to compare the performance difference between the real game scores on each machine and compare then to the expected difference between 3DMark scores. The expected 3DMark overall score is based on the default setting of 3DMark which anyone can download, and the settings Doom3 was designed to run at. Of course that logic is purely my perception and not as accurate as the FPS scores.

[quik_2_win]

Well, that's a difficult comparison. Extrapolating the 'expected performance' correction factor using one openGL and one DX9 application on two totally different platforms employing two different graphics boards can be misleading at best. I do, however understand your logic. It's interesting that you chose GT2 from 3DM03 as I believe it looks very similar to Doom3- look closely at the textures and lighting...however D3 obviously runs much faster. Interesting

[Greg]

It is a difficult comparison. The makers of 3DMark had to use some guess work as to what a future game would look like and what technologies would be used. I notice that Doom3 doesn't use self shadowing characters but 3DMark does. Thief and DeusEx use self shadowing, but no obvious specular lighting. From my and other peoples experience, the CPU plays a significant role in real games but very little in 3DMark. I didn't show a Duron 1.3 with the FX5800 vs a Athlon64 with the FX5800, but I am sure the results would be dramaticaly different.

The only way a better comparison could be made is with a whole bunch of modern games compared relative to 3DMark. My thought was that two years ago, people might have used the 3DMark Battle of Proxycon test to see how their computer might run Doom3 which had been previewed at that time and was much anticipated.

[MUYA]

THE GT2 is dx 8.1 IIRC and uses a whole load of bump mapping...i dunno how much more or less compared to Doom III.

well Doom III is OpenGl game whereas 3Dmark2003 is a DX benchmark. So not entirely comparable. 3dmark is for reference as to "potential" capabilties of a GPU.

I always look for real game benches rather than synthetics when I read review really.

[Greg]

Never mind.

[MUYA]

ya nevermind...wrote that when i was drunk too..

[Woodelf]

Originally Posted by MUYA

ya nevermind...wrote that when i was drunk too..

Been there.