Video Cards - more questions

[Dman3d]

Hello all,

Does anyone know if MAX benefits from SLI?
In stats, when it lists for example, 100 million triangles (polygons?) per second, under what conditions?
I have seen some very nice video cards groan with scenes at 300,000 polys in wireframe, Nvidia 1400 for example. Any shed some light on what the stats mean and why most expensive cards (at least visually in max) seem to behave the same as cheaper/less powerful cards?

see this chart
http://www.nvidia.com/object/IO_11761.html

We typically end up with 50-100+ meg files with between 500k - 1M polys and sometime have files upwards of 2M polys.

[jacksc02]

dman- the scene im currently working on is about 13million polys (of scene geometry) and about 16million polys of proxy information) and my nvidia fx3400 displays everything fine. There is a bit of chug when its refreshing but its great!
Are you talking more for preview animations?

[Mathew Everett]

I am running 2 7800GTX graphic cards in SLI at home and have a Quadro 4400 series at work. and my 7800's stream through projects allot easier than my Quadro card... now if I had 2 Quadro Cards it might be a different story... but in all I do think Max benefits from SLI maybe not 50% or 40% but there is an increase that I have personally seen!

[DaForce]

I was under the impression that SLI made little to no difference in max.
According to this PDF.. it makes a small difference around 10% when using sli quadros
http://www.monarchcomputer.com/Merch...uadroSLIv3.pdf
Not worth it if thats all your using it for. Even in games its iffy

[Dman3d]

Thanks guys. Mainly for moving around in MAX.
I have never seen a card perform smoothly.

jacksc02 -I am stunned that one could handle a scene that big.
That is amazing. Our Quadro 1400 act like it is dying under a scene running 300000 polys. That is when panning, zooming or orbiting. Forget using shaded mode. It is enough to attempt this in wireframe
Do you run in OpenGL, DirectX, Software Z or MaxTreme for MAX?
Also what type of system is it, CPU, RAM and OS?

We have a cross section of cards, wildcat on one (not sure which model have to check with the modeler), Quadros and more aqnd they all perform the same (visually).

I need to move through a scene without the machine dying. They are all Dual 3.02 to 3.06 and 2GB to 4GB Ram.

DaForce - That link is interesting, under MAX 3 at least it looks like the FX3400 is faster than the FX4400 , puzzling, well it looks like single GPU in that test (chart)

[Dman3d]

Found this review

Found this review.

http://www.gamepc.com/labs/view_cont...qfx4400&page=1
It appears at least during this review the SLI where NVidia is concerned is forced off when MAX is loaded, so no gain there. Not sure if this has changes since. I have a 6800 now which makes reading this report a little bit difficult since I am looking at the 4400 or 4500 but I cannot be sure as though they recommend the 4400 for high end big files they seem a little ambiguous about it. See if I can justify the $2k for a video card to the boss.

[DaForce]

See the way i read that PDF the only SLI mode that works within ax is the first one (Multi-view mode) which is only for driving multiple monitors.
So that wouldnt give a performance post as far as im aware.

Odd... there isnt anything thats really definitive.

[jujubee]

Dman you got a 6800 agp? Throw rivatuner on it and softmod it to a quadro 4x series.

[Dman3d]

Have not heard of that one Juju, will look it up asap.

[instinct]

www.guru3d.com

It wont be as fast as a quadro and will prolly be less stable....if you run a 6800 definately use Direct3d, and be sure to use caching in the driver settings....we used to have 6800´s here at work and they performed pretty well using Direct3D. Also keep in mind that everything above 78.xx as nvidia driver SUCKS for openGL.

Regards,
Thorsten

[Dman3d]

I believe that everything over 78 sucks. I grabbed the RivaTune and after much searching and eye poking believe I have it running correctly. At least my machine thinks it is a Quadro 4000, running driver 78.01 but will try 66.93, I think there were some stability issue reported, cant be sure.
It appears to be improving the video (did not have Specviewperf downloaded before I made the modifications) It does not appear to be a lighting change but a couple things confuse me.

The benchmark files at specviewperf for one. There is one file that has 750k polys that flies on this card (post RivaTune) and one of my files that are about the same and less chug to death. Even copying the architectural benchmarks objects several times up to 3 million polys it is moving along nicely. But one of my files of 1/4 the same number of polygons just kills it, in shaded, wire or and combination (visually kills it, clunky and extremely slow) I am guessing this could be where the rest of the system comes into play, not just the card, dunno.

Man this is frustrating. Some files fly while others do not, where with regard to polys, lights and textures they are very similar. Anyone have input on this?

I have never had good luck with direct3d always get bad artifacts on the screen but will give it a try.

[Dman3d]

This the report from SPECviewperf, might help someone.

Dual Xeon 3.02
4GB RAM
GeForce 6800GT 256MB (Using RivaTune to run as Quadro FX4000)
NVidia Driver Set 66.93
Win2k


Run All Summary

---------- SUM_RESULTS\3DSMAX\SUMMARY.TXT
3dsmax-03 Weighted Geometric Mean = 19.58

---------- SUM_RESULTS\CATIA\SUMMARY.TXT
catia-01 Weighted Geometric Mean = 14.25

---------- SUM_RESULTS\ENSIGHT\SUMMARY.TXT
ensight-01 Weighted Geometric Mean = 18.41

---------- SUM_RESULTS\LIGHT\SUMMARY.TXT
light-07 Weighted Geometric Mean = 15.43

---------- SUM_RESULTS\MAYA\SUMMARY.TXT
maya-01 Weighted Geometric Mean = 26.30

---------- SUM_RESULTS\PROE\SUMMARY.TXT
proe-03 Weighted Geometric Mean = 19.84

---------- SUM_RESULTS\SW\SUMMARY.TXT
sw-01 Weighted Geometric Mean = 13.59

---------- SUM_RESULTS\UGS\SUMMARY.TXT
ugs-04 Weighted Geometric Mean = 34.38

According to this list of tests I land in the resoundingly wretched area (with MAX performance, and benchmarks being what they are)
http://www.spec.org/gpc/opc.data/vp81/summary.html

Also see this list
http://www.spec.org/gpc/opc.data/vp81/3dsmax-perf.html

Anyone else out there ran SPECviewperf?

I am going to reset to standard card and re-run the tests and see what happens.

[instinct]

i got no idea about specperfview, but there´s many things that can bog down max. For examples if you are using many groups (esp nested groups)...or many helpers etc etc...also editmesh is a lot faster then editpoly for example. Lots of factors come into play i assume.

Regards,
Thorsten

[RErender]

Here are the results from my fx3400.


Dual Xeon 3.02
2GB RAM
FX3400
NVidia Driver Set 81.95
XPpro

Run All Summary

---------- SUM_RESULTS\3DSMAX\SUMMARY.TXT
3dsmax-03 Weighted Geometric Mean = 26.76

---------- SUM_RESULTS\CATIA\SUMMARY.TXT
catia-01 Weighted Geometric Mean = 20.66

---------- SUM_RESULTS\ENSIGHT\SUMMARY.TXT
ensight-01 Weighted Geometric Mean = 19.03

---------- SUM_RESULTS\LIGHT\SUMMARY.TXT
light-07 Weighted Geometric Mean = 18.87

---------- SUM_RESULTS\MAYA\SUMMARY.TXT
maya-01 Weighted Geometric Mean = 32.86

---------- SUM_RESULTS\PROE\SUMMARY.TXT
proe-03 Weighted Geometric Mean = 35.30

---------- SUM_RESULTS\SW\SUMMARY.TXT
sw-01 Weighted Geometric Mean = 19.68

---------- SUM_RESULTS\UGS\SUMMARY.TXT
ugs-04 Weighted Geometric Mean = 24.58

[Dman3d]

Hmmm, a difference indeed. Thanks for the post.

I reset to defaults ditching the RivaTuner and reset MAX for Direct3D,
much improved (visually on screne in smooth/hilight without edged faces, with edged faces it chugs but nowhere near as bad as when I first made this post).
Will run the benchmark again tomorrow and see what happens
wiuth 81.xx tomorrow and see if the DirectX relationship is the same.
I wonder about the above mentioned issues with OpenGL.
Many of our workstations have the lower end Quadro's. If they are being
crippled by the driver that will be disappointing.

Eric, what driver to you use for MAX, MAXtreme?