Best Video card for the best viewport perfomance

[Nizar]

I know was debated before, but now there is new VGA released by Nvidia and AMD and only now I need to upgrade my card...

So, anyone has advice about a video card to buy? I'm oriented on AMD gaming card.

[doca]

Hi. As it is mentioned in some forums, nVidia is forcing you to get professional quadro cards by degrading some performances (from XSI user pov-degrading openGL performance). How they do that, I dont know, but I can tell you my experience few days before. I have tested nVidia gtx 680, nVidia gtx 570 and 4 of 5 years old nVidia gtx 285. I have them all, currently in my working machine 680, and in my secondary machines later two gtx-s. Even now, 285 is outperforming 570 and 680 in viewport performances. Not too much, about 3-5% in cinebench, but that is about 5 years old card. Maybe even more, cant remember. Also, seems to me that working in xsi is more comfortable with 285.
I can speak only about nVidia graphics, as I dont have ATI. Currently, it is very hard to find well balanced gaming card for work, and I am very unhappy with gtx680.
But, at the end, why dont we make one xsi test scene, to get results from various cards, and to find out what is the ultimative XSI graphic card, both professional and gaming cards.

[rray]

Code: Select all

ImportModel("C:\\Program Files\\Autodesk\\Softimage 2013 SP1\\Data\\XSI_SAMPLES\\Models\\Characters\\XSI_Man_Armored.emdl", null, null, null, null, null, null);
SelectObj("XSI_Man_Armored.Body", null, true);
ApplyTopoOp("MeshLocalSubdivision", "XSI_Man_Armored.Body.poly[*]", 3, siPersistentOperation, null);
SetValue("XSI_Man_Armored.Body.polymsh.meshlocalsubdivisionop.subdivisionlevel", 3, null);
SetValue("Camera.camvis.refreshrate", true, null);
SelectObj("XSI_Man_Armored.Body", null, true);
ActivateRaycastPolySelTool(null);
SetDisplayMode("Camera", "textured");
SelectAllUsingFilter("Polygon", siCheckComponentVisibility, null, null);

GTX 670 (Asus DirectCU2): (Really happy with the card although it at first crashed SI. Worked after patching a dll.)
15 fps unsubdivided (1 mil polygons) with all polygons selected (subcomponent highlighting which draws very slow in softimage)
60 fps unsubdivided without selection
2 fps subdivided (4 mil polygons) all polygons selected
45 fps subdivided, when all objects are unselected

GTX 460
10 fps unsubdivided, all polys selected
60 fps unsubdivided, no selection
1 fps subdivided, all polys selected
30 fps subdivided, no selection



*edit: sorry should have clarified - framerate is with with full model in view, and viewport orbiting, so the envelope doesn't have an influence

[doca]

Thanks rray! Your scene is as script opens it, with only legs visible, isn't it? And only hit "play", just to be clear?
Can I ask, what dll you patched and why?

Gigabyte GTX 680/ intel i7-930
12.5 fps
23 fps

3.2 fps
8.5 fps

Gainward GTX 285/ intel i7-2600k
12.5 fps
20.5 fps

3.7 fps
9 fps

[sonictk]

Since we're comparing our d**k sizes here Just wanted to pipe in and say that Quadro isn't always necessarily better...but I would stick with NVIDIA because of CUDA and PhysX...

Quadro k2000M (i7 3630 QM, 24 GB RAM)

~15 FPS (no subD, unselected)
~10 FPS (no subD, selected)

~ 0.2 FPS (subD lv 1, unselected)
~ 0.1 FPS (subD lv1, all selected) (actually much slower, but 0.1 is the lowest the readout will go )

Worth mentioning: my GTX 560Ti on my main rig (i5-2500k, 16 GB RAM) actually outperforms the Quadro by quite a bit (almost a factor of two), which I think is hilarious, all things considered.

However, that being said, I can actually paint with my Quadro in Mudbox at full texture resolution, on multiple 16bpp layers with zero/little slowdown, whereas on my GeForce, everything slows down to a crawl after a certain point and I am forced to merge my layers in order to get usable performance. Perhaps it's also on the part of the XSI viewport in dire need of some performance improvements as well...

I dunno, I was expecting like a MILLION FPS when I bought this thing, but alas. Also i should mention that swapping between subD levels took like an supremely-long amount of time as well, and switching beyond subD 2 for the man actually crashed XSI.

I should import one of my old environment scenes and have a HQV test of that.

[ActionArt]

I remember this cropping up on the the xsibase forum. There's one big question here:

Why is rray's 670 way faster than doca's 680?

That makes no sense. Are the top end cards hobbled more than low end cards? Is there something else wrong? Is there a setting that makes the difference? This is a mystery...

I remember in xsibase my 470 way outperformed a 480 at the time. Same issue.

[mirkoj]

you have influence of cpu in this scene as well with all envelopes to be processed. there is older scene that more suits pure GPU performance.
actualy I think that was back on xsibased... here is script:

Application.CreatePrim("Cube", "MeshSurface", "", "")
Application.SetValue("cube.polymsh.geom.subdivu", 831, "")
Application.SetValue("cube.polymsh.geom.subdivv", 800, "")
Application.SetValue("cube.polymsh.geom.subdivbase", 800, "")
Application.SetValue("Camera.camvis.refreshrate", True, "")
Application.SetDisplayMode("Camera", "shaded")
Application.DeselectAll()
Application.SetValue("PlayControl.Out", 5000, "")
Application.DeselectAll()
Application.GetPrim("Null", "", "", "")
Application.SelectObj("Camera_Root", "", "")
Application.CopyPaste("Camera_Root", "", "null", 1)
Application.SelectObj("null", "", "")
Application.SaveKey("null.kine.local.rotx,null.kine.local.roty,null.kine.local.rotz", 1, "", "", "", "", "")
Application.SetValue("PlayControl.Key", 5000, "")
Application.SetValue("PlayControl.Current", 5000, "")
Application.Rotate("", 0, 8000, 0, "siAbsolute", "siPivot", "siObj", "siY", "", "", "", "", "", "", "", 0, "")
Application.SaveKey("null.kine.local.rotx,null.kine.local.roty,null.kine.local.rotz", 5000, "", "", "", "", "")
Application.FirstFrame()

And results with 7970:
Shaded: 160 fps
Wire: 120 fps

With some older version of drivers I had up to 175 fps in shaded and with just a bit OC can easily push to 200 fps in shaded.
Problem with gtx580 was that once started play first it would stick at around 17 fps for 30-40sec and only after that it could up to 60-70 if I remember correctly. WOuld be great to see updated results with some of never cards as well.

[rray]

you have influence of cpu in this scene as well with all envelopes to be processed. there is older scene that more suits pure GPU performance.
actualy I think that was back on xsibased... here is script:

Thanks rray! Your scene is as script opens it, with only legs visible, isn't it? And only hit "play", just to be clear? Can I ask, what dll you patched and why?

Sorry I meant orbiting because with play the envelope does have an influence. I didn't use the cube scene because I looked for something with a texture. I patched mentalray.dll, Softimage crashed on startup, before showing the viewports, without that patch.

To get back more on topic, can anybody test this with an ATI?

[mirkoj]

now it is more clear )

ati hd7970

orbiting with everything visible in view unselected: ~88 fps
geometry selected object mode: ~60 fps
geometry component mode: 8.6 fps

subdiv (once right?)
orbiting with everything visible in view unselected: ~82 fps
geometry selected object mode: ~30 fps
geometry component mode: 0.8 fps

body head and gun from geometry selected.

[Nizar]

So, seems ATI 7970 wins these round. Actually the price seems nice, 350 € for a model with 3gb

A shame these card are not useful for gpu rendering.

[mirkoj]

yea shame that new Redshift renderer is still not supporting opencl.
When and hopefully it will once it will be no brainer to think what will be your next card for SI.

btw would be grat if someone with some firepro series card do this test as well. Still wondering if there would be some performance gain
with for example w7000. That one seems to have pretty nice price but would it be faster than 7970?

[origin]

I have selection lag with ati 7870...

[Nizar]

I have selection lag with ati 7870...

Not a good news...

I don't use at all HQV, but how ATI will perform under HQV environment?

[ActionArt]

yea shame that new Redshift renderer is still not supporting opencl.

I don't think it's too fair to expect that at this point. Developing for 1 platform would be hard enough and from what Luceric and others have said, opencl is a mess right now, way behind cuda. So until that gets straightened out, I wouldn't expect to see it.

[Nizar]

yes but a quadro look like very too expensive to be honest...

[ActionArt]

Using rray's script in a GTX 470 I got these results:

GTX 470
18 fps unsubdivided, all polys selected
24 fps unsubdivided, no selection
6 fps subdivided, all polys selected
14 fps subdivided, no selection

Then a strange thing happened. After I subdivided and tested, I un-subdivided and it went from 24fpx to 45fps!

Results are very inconsistent for all modes. It's all over the board depending on what you do first.