3dmark05 image quality analysis?

[thegrommit]

Does the single frame rendering option produce the same scene across different cards?

These are all supposed to be frame 327 of the firefly game test:

6800 Ultra - taken using the single frame rendering option.

6800GT - taken during a benchmark run (ignore the artifacts - note how it's a different scene).

X800 Pro frame 328 - as close as I could get on a benchmark run (I don't have the Pro version).

 

[carpediem]

Does the single frame rendering option produce the same scene across different cards?

These are all supposed to be frame 327 of the firefly game test:

6800 Ultra - taken using the single frame rendering option.

6800GT - taken during a benchmark run (ignore the artifacts - note how it's a different scene).

X800 Pro frame 328 - as close as I could get on a benchmark run (I don't have the Pro version).

I guess it's more accurate on the pro edition, where you can choose what frame to render

Maybe the frame in the actual benchmark depends on the frame rate?

 

[Scali]

I noticed that too, on certain sites that tried to compare the same image on different cards. I guess it's because it doesn't try to render all frames on every card, but it tries to render the same animation sequence.
So, the tests take equally long on all hardware, but the faster the hardware, the more intermediate frames are rendered (which is a good thing, else it'd take days to benchmark an FX5200 ).
So I think the framecounter in benchmark mode is simply counting how many frames it rendered. I suppose at the end of each game test the differences in the framecounter should be the largest then.
Notice the difference in time here, at the same frame. You should try to make a screenshot at the exact same time rather than at the same framecount.

 

[thegrommit]

I guess it's more accurate on the pro edition, where you can choose what frame to render

Right - I'm hoping someone with a X800 (of any sort) and the Pro edition can confirm that the same scene is being rendered as on the 6800U.

Scali - that kinda makes sense. Hmmn. Let me see how many frames are rendered for the first game test.

[edit] Approximately 2188 frames in 1:42.77. If Scali's theory is correct, the total number of frames on (say) a 9800 Pro should be substantially less.

 

[thegrommit]

Page 18 of the whitepaper contains the explanation under Image Quality Tool:

When generating the frames, the tool uses frame-based rendering at 30 frames per second.

 

[DOGMA1138]

well heres frame 327 on my x800 card. at least i can feel usefull :?
http://planet.nana.co.il/zvizka/IQ1.jpg

 

[Rys]

If you think about it logically, even without access to the actual Image Quality tool , surely it would be fairly stupid for FM to include the tool if it didn't output exactly the same frame each time you asked it to?

Rys

 

[thegrommit]

If you think about it logically, even without access to the actual Image Quality tool , surely it would be fairly stupid for FM to include the tool if it didn't output exactly the same frame each time you asked it to?

Rys

It seemed a reasonable question to ask given that the "same" frame was so different between the X800 and 6800. I wasn't sure whether the tool was only intended for comparing image quality between different drivers and the refrast on the same hardware.

Scali was kind enough to provide an (correct) explanation. What was the purpose of your post?

 

[Rys]

Scali was kind enough to provide an (correct) explanation. What was the purpose of your post?

My apologies if you took offence, it's just this is not the first thread that's asked the same, in my eyes, redundant question about the IQ tool.

Rys

 

[Rolf N]

If you think about it logically, even without access to the actual Image Quality tool , surely it would be fairly stupid for FM to include the tool if it didn't output exactly the same frame each time you asked it to?

Rys

Yes. There was such an issue with 3DMark03, IIRC some "procedural" geometry like hair weren't deterministic. I even remember that there was quite a fuss about it.

If my memory isn't playing me any tricks, this would be reason enough to ask this question irt 3DMark05.