Do timedemos accurately predict real-world performance?

[imported_SolarWind]

I've noticed that I tend to get much higher fps in UMark than when actually playing through similar situations in the same level with the same image quality settings. What information normally computed by the CPU in UT2k4 or Doom 3 is obtained from the timedemo file? Assuming the timedemo is a recording of typical gameplay, do timedemo benchmarks tend to give the correct performance ranking and percentage performance difference for video cards and CPUs even though they raw fps numbers are much higher than in actual gameplay? I would suspect that timedemos at least give the correct performance distribution for video cards since even if the CPU load is lighter than normal, the video card still has to do the same work per frame. However, wouldn't differences in the CPU work pattern and load when running a timedemo make the results suspect for comparing CPUs? performance in actual gameplay?

Also, I've noticed that when using demorec in UT2k4, demoplay with sound enabled only gives me about 1/2 to 2/3 the fps as I was getting when I recorded the demo.

 

[Tab]

I'd say yes, it is infact the actual game engine. Its not going to be 100% accurrate but its better than 3dmark. How accurrate the time demo is based upon if the game designers show you what the game is going to require your computer to do.

 

[Concillian]

I would say:
For VIDEO CARD performance: YES
For CPU performance: NO

CPU usage is lower in demos for most games.

 

[Dman877]

XBit has an article up on that right now actually. Apparantly Intel actually runs Doom3 better then AMD.

 

[BFG10K]

It depends on the demo. If it's running in real time and doing exactly what it would be doing if you were actually playing that section of the game then absolutely.

 

[Jeff7181]

In a well recorded timedemo, yes. The Doom 3 timedemo does. But almost any timedemo in Doom 3 would because it's much of the same over and over. Far Cry on the other hand would definately need two timedemo's to give you an accurate indication of performance. One timedemo for inside levels, and one for outside... the differences are drastric, and if you give one or the other, or an average of both, it won't represent the game as a whole.

 

[BFG10K]

The other problem with Far Cry's demos is that they don't really test physics or AI as the monters just follow their standard patrol paths.

 

[Jeff7181]

Originally posted by: BFG10K
The other problem with Far Cry's demos is that they don't really test physics or AI as the monters just follow their standard patrol paths.

True.