View Full Version : 4x SSAA Ti4600 scores
Ti4600@317/729, KT400, 512 MB PC2100, 1900+
3D Mark 2001 SE b330 default settings:
0x FSAA 10360
2x MSAA 8609
4x MSAA 6722
4x SSAA 5639
In game 1 and 3 low detail tests I get 70-120 fps with 4x SSAA.
Just kind of a followup to the R300+SSAA discussion where some felt the 9700 Pro won't have the performance to do it.
Ok so the 9700Pro only has 64 Mtex higher fillrate (^^ OCed the Ti4600).
But almost twice the bandwidth, better HSR/compression and higher pixel fillrate would surely grant it a score several hundered marks higher than this when using SSAA IMHO.
Any thoughts?
(BTW as a reference I get
0x FSAA 12880
2x MSAA 11106
4x MSAA 9550
on my 9700 Pro non OCed in the same rig, but those scores are using pretty old drivers and FSAA got a boost in the cat 2.5 or was it 3.0)
Now gimme my SSAA :D
Bambers
19-Jan-2003, 11:10
You sure thats 4xSSAA?
I get 2500 on my 8500 (315/315) with 4xSSAA. With 2xSSAA I get about 5000 and 70-100fps in the low detail tests. You sure it isnt running 4xS or something?
You sure thats 4xSSAA?
I get 2500 on my 8500 (315/315) with 4xSSAA. With 2xSSAA I get about 5000 and 70-100fps in the low detail tests. You sure it isnt running 4xS or something?
I think so, I'm using the 4x Super sampling mode in "aTuner!" which according to 3dcenter.de is pure 2x2x SSAA
by the looks of it, it surely is SSAA and edge AA quality is a bit better than 2x MSAA
so I think it's pretty safe to say that it's 4x SSAA, but of course I can't be 100% sure
Colourless
19-Jan-2003, 14:24
Um, wouldn't that be the 4xS mode with 2x Supersampling and 2x Multisampling?
Bambers
19-Jan-2003, 14:46
Do the fillrate tests. The scores should be 1/4 of no AA for 4xSSAA. Should be around 250 and 500 for single and multitexture. If they're 500 and 1000 then it's probably 4xS.
Do the fillrate tests. The scores should be 1/4 of no AA for 4xSSAA. Should be around 250 and 500 for single and multitexture. If they're 500 and 1000 then it's probably 4xS.
hehe I got something in between ;)
I get ~1100/2400 without FSAA and ~320/1000 with 4x "SSAA"
here's what 3dcenter says about the new mode:
"nVidia 4xSS anti- Aliasing mode: To implement very simply to find therefore frequently with most diverse diagram chips: 2x2 Supersampling. However relatively little edge smoothing offers for the necessary achievement."
"The 4xSS-Modus is most uninteresting. It concerns here pure 2x2 Supersampling, as it is available on GeForce 1/2 until today. We do not recognize so quite the sense, since 8xS smoothes very much better, but only approx.. 30 per cent more achievement costs. But rather a mode too much than too few: Nostalgikern and Supersampling fan are now again to 4x Supersampling at the disposal."
"4xSS supplies better texture quality opposite 4x a something, is however very many slower. Textures one polishes besides better with anisotropic filtering up."
Accordign to the it's straight 2x2x OGSS
from the looks of it I'd agree but perhaps the edge AA
it certainly isn't 4xS since texture quality is higher and the sample pattern is obviously different
nor does it look identical to any of the other AA modes since it offers the highest texture quality of them all
take a look here:
http://www.nordichardware.se/artiklar/Grafik/2003/nv25AA/
I tried searching for that FSAA tester which one of the members here made but I can't remember what it's called, perhaps it could be more usefull when trying to find out what this mode is
Here you go: link (http://hem.passagen.se/basic3/fora/beyond3d/fsaaview.htm)
Dave Baumann
19-Jan-2003, 16:12
Basic, I know its a pain in the ass, but have you got the time to update v0.3 and make it available for D3D? The Geomtry mode would be very handy for detecting the moded mode NV are using for D3D only.
Bambers
19-Jan-2003, 16:17
Your multitexture fillrate seems too high. Would have thought it should be around 600. Unless its dropping back to 4xS or something for 3dmark :?
I did some tests with those new AA-modes using my AA sample position test program.
Apparently, all new AA-modes in the aTuner (4X SS, 6X and 8X) all are ordered grid, where 4X is 2x2, 6X is 3x2 (where four of them has 1/8 weight and two of them has 1/4 weight), and 8X is 4x2.
By the way, 8XS is a 2x2 grid of 2X AA.
Your multitexture fillrate seems too high. Would have thought it should be around 600. Unless its dropping back to 4xS or something for 3dmark :?
yeah I know, that's why I'm so curious about this mode and its performance :)
here's a pic from FSAAview 0.2:
http://www.nordichardware.se/artiklar/Grafik/2003/nv25AA/4xSSview.png
Strange... I tested with my fillrate program on a GF4 Ti 4200. Using 2 textures, I got 980Mpix/s for no AA, and only 148Mpix/s for 4X SSAA. Using 4 textures, I got 491Mpix/s for no AA and only 93Mpix/s for 4X SSAA.
Perhaps there are really something strange with 3DMark...
Strange... I tested with my fillrate program on a GF4 Ti 4200. Using 2 textures, I got 980Mpix/s for no AA, and only 148Mpix/s for 4X SSAA. Using 4 textures, I got 491Mpix/s for no AA and only 93Mpix/s for 4X SSAA.
Perhaps there are really something strange with 3DMark...
3d mark measures texel fillrate, not pixel
could you send me that proggy that measured pixel fillrate so I can try it out?
It is easy to convert: 980Mpix/s with 2 textures -> 1960Mtexels/s, and 148Mpix/s with 2 textures -> 296Mtexels/s. 491Mpix/s with 4 textures -> 1964Mtexels/s, and 93Mpix/s with 4 textures -> 372Mtexels/s.
My fillrate test program can be found here (http://www.csie.ntu.edu.tw/~r89004/fillrate032.zip). Check the config.txt file in the zip file first. I got the test results under 1024x768 16 bits mode.
If there were any question about it: Yes, that's straight 2x2 OGSS.
Dave:
You're right about the first part. :shock: :D
It's put on hold without any plan to change the status. The reason is that D3D and I don't like each other. If somebody can tell me how to do the equivalent of a glBegin(GL_TRIANGLE_STRIP); glVertex3f(...); ... glEnd(); - sequence in D3D, then I might do it. I don't want to keep binding vertex arrays, or make one big mesh, just send vertices in realy immediate mode. I probably missed some simple functions.
Colourless
19-Jan-2003, 18:55
I think I'll add, were you using Texture Mode, or Geometry Mode with FSAAView. Remember that MSAA has no effect on Texture Mode.
Colourless
19-Jan-2003, 19:00
It's put on hold without any plan to change the status. The reason is that D3D and I don't like each other. If somebody can tell me how to do the equivalent of a glBegin(GL_TRIANGLE_STRIP); glVertex3f(...); ... glEnd(); - sequence in D3D, then I might do it. I don't want to keep binding vertex arrays, or make one big mesh, just send vertices in realy immediate mode. I probably missed some simple functions.
There isn't any way, unless you were to duplicate the functionality of those yourself. I.E. create functions where glBegin() locks the Vertex Buffer, with glVertex3f() adds a vertex and increments a pointer, and glEnd() locks the buffer and draws it.
The alternate was is to just use IDirect3DDevice8::DrawIndexedPrimitive()
Colourless 1:
FSAAview 0.2 only has the texture mode.
Colourless 2:
Thanks for the info. I guess I have to keep locking vertex buffers. I just wanted to treat each poly in an 100% identical way to make sure that all operations are done in the same way (since the algorithm can be sensitive to rounding errors). And since it's most convenient I'd rather have a simple DrawQuad() in the innermost loop. Making a big vertex array isn't fun since it could be well over 100MB.
Is locking vertex buffers a couple of million times for one frame gonna be slow? :?
Yes, I know I'm whining over a small and insignificant problem. It's just that if I see that something will be ugly whatever I do, I can get this uncontrollable disgust to the problem. :)
Soooo... how the hell do my card get these results which by all means, except the performance, seems to be 2x2x OGSS? ;)
Anyways though this mode is great for some games that I play I'd still want it in Open GL..
I hope the FX has some SSAA modes enabled.
Bambers
19-Jan-2003, 20:09
Take a screenshot in one of the game tests where you were getting 70+fps from 3dmark, then you can tell if it is OGSS or if somehow 3dmark is causing another mode to be used.
Colourless
19-Jan-2003, 20:37
Basic locking a vertex buffer that manytimes, might be just a 'tad' slow. However, it might not matter too much, if you are only drawing the frame once.
Anyway, would it be possible to create a single vertex buffer of say 20k verts and then use the World matrix to just shift the transformed positions of the verts, sort of like how you can use a display list in OpenGL?
You're right, it isn't like it need to get a high framerate. :) And "a couple of million" is actually 2M for 1600x1200, and less for smaller windows. It's one quad per pixel IIRC.
Shifting a mesh has been concidered. But I didn't like it because it would mean a non-uniform way of setting the position of polys. Within the mesh, it would be the function filling the vertex array that makes a difference. On mesh borders, it's the shift in the world matrix.
In this test I'm intentionally paranoid about having all rounding errors as uniform as possible. Just because it's rather sensitive here. It's probably overkill, but otherwise I would always think about if unexpected stuff could be mesh-border-errors.
Others:
Sorry for the thread-hijacking. Just move along past us, nothing to see here.
Take a screenshot in one of the game tests where you were getting 70+fps from 3dmark, then you can tell if it is OGSS or if somehow 3dmark is causing another mode to be used.
How do I grab screenies in 3D Mark?
I've never been able to.
I've tried the basic print screen, HyperSnapDX and Fraps, none of them worked.
OpenGL guy
20-Jan-2003, 01:32
How do I grab screenies in 3D Mark?
I've never been able to.
I've tried the basic print screen, HyperSnapDX and Fraps, none of them worked.
I believe F12 will do the trick.
How do I grab screenies in 3D Mark?
I've never been able to.
I've tried the basic print screen, HyperSnapDX and Fraps, none of them worked.
I believe F12 will do the trick.
ahh thanks, they shoudl really be more clear on that since Print Screen and other screen grabber proggies doesn't work with their software
here's the pic, personally I think it looks like expected(?)
http://www.nordichardware.se/artiklar/Grafik/2003/nv25AA/max4xss.png
AdamK47
20-Jan-2003, 06:32
Does this work in OpenGL with these drivers? It's supersampling alone, so I should work. Is there a hidden option implemented for OpenGL.
AdamK47
21-Jan-2003, 18:02
Oh great! I've imposed a question that has both ended this thread and destroyed the minds of all who read it by the shear complexity of it all.
Does this work in OpenGL with these drivers? It's supersampling alone, so I should work. Is there a hidden option implemented for OpenGL.
No unfortunately all the "hidden" modes are only availible in DX.
Guess you'd have to use NV1x emulation to get SS in OGL. :(
here's the pic, personally I think it looks like expected(?)
http://www.nordichardware.se/artiklar/Grafik/2003/nv25AA/max4xss.png
That's definately 4x OG AA.
*wonder what his 9700 Pro would score with 4x SSAA*
guess I'll never know since ATi doesn't seem to read their own product presentations ;)
vBulletin® v3.8.6, Copyright ©2000-2013, Jelsoft Enterprises Ltd.