Page 5 of 8

Posted: Tue Sep 18, 2007 11:07 am
by Fiesch
hehe ok from those posts i can spare myself the trouble of replacing the graphics card in my workstation... would have been a 6600GT ;)

Posted: Tue Sep 18, 2007 11:15 am
by Fiesch
Hm ok the message sending would be problematic, the log has 20 mB ;) my server blocked that for now - i could open the filter for that on my sie but i don't know if your account would accept that

Posted: Tue Sep 18, 2007 11:18 am
by Oogst
Does that mean it actually started and showed something? The log should only be that large if you ran the entire benchmark with it or something like that, because if you just start and close after a few seconds, it should be much smaller.

Posted: Tue Sep 18, 2007 11:23 am
by Fiesch
it ran all tests through ,some were just plain white - but it did execute all benchmark cases

Posted: Tue Sep 18, 2007 11:32 am
by Oogst
Ah, nice! In that case, I do not need the debugging information, because it is already fixed. :) I have uploaded a new version of the debugging thing now, which solves the problem without creating so much log. Also, it mention that it should be run at 32 bits. :)

I you could download again (same link as before) and run it on your ATI, then that what be really nice!

Posted: Tue Sep 18, 2007 11:33 am
by Devil N
And another test result for mr. Oogst.

Posted: Tue Sep 18, 2007 11:45 am
by Oogst
Thanks!

Funny, your computer is almost exactly the same as mine! Only your processor is slightly faster than mine. Test-results are almost exactly the same as well.

Posted: Tue Sep 18, 2007 12:08 pm
by Devil N
Good, that means I didn't screw up ;)

Posted: Tue Sep 18, 2007 12:08 pm
by Fiesch
i've sent my data out now, too

Posted: Tue Sep 18, 2007 2:41 pm
by mr.Zog
Benchmarked with two PCs! (see mail)

Posted: Tue Sep 18, 2007 3:34 pm
by Shadow007
Well done Oogst !!

Just two remarks :
There seem to be some artifacts : some guys are cut in half !
would it be possible to put them more "in the depth" of the rooms rather than at the window ?

Posted: Tue Sep 18, 2007 4:45 pm
by Chris H
Very cool effect, Oogst. I have a question: have you compared the speed of this method to the speed of using regular geometry?

Posted: Tue Sep 18, 2007 4:52 pm
by Fiesch
That's what the benchmark does ;)

Posted: Tue Sep 18, 2007 4:54 pm
by Chris H
Then which is faster? The complex shader on a cube, or the simpler shader on more complex geometry?

Posted: Tue Sep 18, 2007 5:45 pm
by Lothar
I sent my data too... hope it'll be usefull 8-)

Posted: Tue Sep 18, 2007 6:12 pm
by jjp
Chris H wrote:Then which is faster? The complex shader on a cube, or the simpler shader on more complex geometry?
With a low number of meshes there is hardly any difference. The more meshes and polygons you get the better interior mapping looks. Playing around with the benchmark e.g. at some point it was like 30fps with real geometry and close to 200fps with interior mapping.

In practice I think one good application would be a flight simulator where this could make displaying cities a lot cheaper.

Posted: Tue Sep 18, 2007 7:39 pm
by Oogst
Thank you, Lothar, mr.Zog, Nikolaj and Antonio for sending in your results!
Shadow007 wrote:Well done Oogst !!

Just two remarks :
There seem to be some artifacts : some guys are cut in half !
would it be possible to put them more "in the depth" of the rooms rather than at the window ?
Yeah, with the guys it suddenly becomes very important to exactly tile and position them. Not really a problem, but I did not put any effort into it. As for the cutting into half at the corners of the building: that is because unlike the interior walls and floors, the characters are positioned parallel to the exterior wall. So if you look at them through a polygon with a different orientation, they are gone. A bit odd, I know, but it is easily solved by not making windows that overlap the corners of the building. The material with windows and characters hides the seams through the characters at the corners of the building behind a wall. :)
jjp wrote:
Chris H wrote:Then which is faster? The complex shader on a cube, or the simpler shader on more complex geometry?
With a low number of meshes there is hardly any difference. The more meshes and polygons you get the better interior mapping looks. Playing around with the benchmark e.g. at some point it was like 30fps with real geometry and close to 200fps with interior mapping.

In practice I think one good application would be a flight simulator where this could make displaying cities a lot cheaper.
The point of the performance is that Interior Mapping is exactly as expensive as the number of pixels on the screen it occupies. Using early z-out, Interior Mapping is not calculated on objects behind the first object, so adding more objects is not a problem. With normal buildings, the bottleneck is the complexity of the building and the number of render calls (more are needed to be able to give the floors, ceilings and walls seperate tiling textures), so adding extra buildings that are behind other buildings is just expensive as when they were fully in view.

Interior Mapping would by the way also have been awesome in Superman, Crackdown (which claims to do something like it, but does not show it at all) and Spiderman. All have lots of high buildings with lots of windows.

Posted: Tue Sep 18, 2007 7:46 pm
by Oogst
Tomorrow I will post an XLS-file with all the results I got up to then. Does anyone mind if his name is in that, or should I just remove all names and just leave the hardware information there?

Posted: Tue Sep 18, 2007 9:28 pm
by PolyVox
Glad you're getting the results you need! And personally I don't mind my name being included.

Posted: Tue Sep 18, 2007 10:53 pm
by Chris H
I tried to do a benchmark, but I guess this doesn't support the shaders. I won't submit my results because they're not likely to be of any use to you.

WARNING: material IM_FullyTexturedRooms_AlphaPlane_LightVariation has no supportable Techniques and will be blank. Explanation:
Pass 0: Fragment program Cg_IM_FullyTexturedRooms_AlphaPlane_LightVariation_FP cannot be used - not supported.

Posted: Wed Sep 19, 2007 9:12 am
by mr.Zog
PolyVox wrote:And personally I don't mind my name being included.
Same here ;)

Posted: Wed Sep 19, 2007 9:42 am
by Fiesch
would be more "scientific" to exclude names, but i don't really mind either

Posted: Wed Sep 19, 2007 12:48 pm
by Cobra8472
Oogst wrote:Thank you, Lothar, mr.Zog, Nikolaj and Antonio for sending in your results!
Shadow007 wrote:Well done Oogst !!

Just two remarks :
There seem to be some artifacts : some guys are cut in half !
would it be possible to put them more "in the depth" of the rooms rather than at the window ?
Yeah, with the guys it suddenly becomes very important to exactly tile and position them. Not really a problem, but I did not put any effort into it. As for the cutting into half at the corners of the building: that is because unlike the interior walls and floors, the characters are positioned parallel to the exterior wall. So if you look at them through a polygon with a different orientation, they are gone. A bit odd, I know, but it is easily solved by not making windows that overlap the corners of the building. The material with windows and characters hides the seams through the characters at the corners of the building behind a wall. :)
jjp wrote:
Chris H wrote:Then which is faster? The complex shader on a cube, or the simpler shader on more complex geometry?
With a low number of meshes there is hardly any difference. The more meshes and polygons you get the better interior mapping looks. Playing around with the benchmark e.g. at some point it was like 30fps with real geometry and close to 200fps with interior mapping.

In practice I think one good application would be a flight simulator where this could make displaying cities a lot cheaper.
The point of the performance is that Interior Mapping is exactly as expensive as the number of pixels on the screen it occupies. Using early z-out, Interior Mapping is not calculated on objects behind the first object, so adding more objects is not a problem. With normal buildings, the bottleneck is the complexity of the building and the number of render calls (more are needed to be able to give the floors, ceilings and walls seperate tiling textures), so adding extra buildings that are behind other buildings is just expensive as when they were fully in view.

Interior Mapping would by the way also have been awesome in Superman, Crackdown (which claims to do something like it, but does not show it at all) and Spiderman. All have lots of high buildings with lots of windows.
You are incorrect regarding Crackdown. If you jump to a building and look at the windows you can clearly see the cieling/walls inside.

It's quite an awesome effect and really adds to the immersion factor.

Posted: Wed Sep 19, 2007 3:14 pm
by Oogst
Cobra8472 wrote:...

You are incorrect regarding Crackdown. If you jump to a building and look at the windows you can clearly see the cieling/walls inside.

It's quite an awesome effect and really adds to the immersion factor.
Hmm, interesting! I have not been able to find that effect on any footage of the game, though. :(
Fiesch wrote:would be more "scientific" to exclude names, but i don't really mind either
They won't be mentioned in the paper. It is just easier if I upload stuff especially for you folks, because you do not have to try to figure out which is your own PC. :)
Chris H wrote:I tried to do a benchmark, but I guess this doesn't support the shaders. I won't submit my results because they're not likely to be of any use to you.

WARNING: material IM_FullyTexturedRooms_AlphaPlane_LightVariation has no supportable Techniques and will be blank. Explanation:
Pass 0: Fragment program Cg_IM_FullyTexturedRooms_AlphaPlane_LightVariation_FP cannot be used - not supported.
That probably only goes for 2 of the 97 tests, because the rest is shader model 2.0 and these two are shader model 2.x. I would actually be interested in your test-results, because your GPU is of an older generation than the rest, which is interesting for diversity. I can simply exclude these two specific tests from the test results.

Posted: Wed Sep 19, 2007 3:26 pm
by Cobra8472
Oogst wrote:
Cobra8472 wrote:...

You are incorrect regarding Crackdown. If you jump to a building and look at the windows you can clearly see the cieling/walls inside.

It's quite an awesome effect and really adds to the immersion factor.
Hmm, interesting! I have not been able to find that effect on any footage of the game, though. :(
Fiesch wrote:would be more "scientific" to exclude names, but i don't really mind either
They won't be mentioned in the paper. It is just easier if I upload stuff especially for you folks, because you do not have to try to figure out which is your own PC. :)
Chris H wrote:I tried to do a benchmark, but I guess this doesn't support the shaders. I won't submit my results because they're not likely to be of any use to you.

WARNING: material IM_FullyTexturedRooms_AlphaPlane_LightVariation has no supportable Techniques and will be blank. Explanation:
Pass 0: Fragment program Cg_IM_FullyTexturedRooms_AlphaPlane_LightVariation_FP cannot be used - not supported.
That probably only goes for 2 of the 97 tests, because the rest is shader model 2.0 and these two are shader model 2.x. I would actually be interested in your test-results, because your GPU is of an older generation than the rest, which is interesting for diversity. I can simply exclude these two specific tests from the test results.
Yeah, I tried to find one to illustrate, but no luck. I'll pull out my digicam and snap a pic when I can for ye :)

It looks very similar to yours, except that it kind of fades into the void (fog?).

Definately is awesome when you're jumping from windowsill to windowsill and actually see something inside, instead of a black texture.

Good work!