Doomtrooper
Veteran
Thanks Galilee..now the 30.82 drivers..you should see a huge peformance increase since the Pixel Shader scores on 3Dmark showed a 30-40% gain in performance on Nature...
nVidia release new all singing all dancing dets.
- Thread starter BoardBonobo
- Start date
Doomtrooper
Veteran
Thx 
Hmmm I wasnt able to reproduce Galilee's findings unfortunately. But then again, I tested at 1280x1024x32 with the 40mb textures and pixel shaders ON for all tests.
Definately need more testing to try and see I can get a configuration, registry or driver set to match Galilee's findings. Will keep trying to find the missing link, which admittedly my set of tests are half hearted at best on a "in use" PC which could be the culprit.
....
Seems to be the case with 3DMark and Aquamark at least. Not sure if this is always the case as more benchmarks would be needed.
Well, one could call 8x which is precisely 4x "just fine" since it indeed runs just fine. As for backing up the claim, just flip back a couple pages in this thread with the custom level UT shots you guys provided. There *is* a performance difference between the identical 4x and 8x shots, just benchmark. The main bit of magic in the 40.xx drivers is that under some conditions, the nearly identical 4x and 8x dont show much of a performance hit (ala Aquamark) but do if you ramp up the textures or use a benchmark that isnt overly cpu/poly bound. (see the little charts as provided of 3dmark tests that show this trend quite plainly) you will see a difference in performance between 4x and 8x, so it cant simply be swept under the rug as "well, 8x just isnt working" since it will produce a different final score in, say, 3dmark?
As has already been quoted on this very thread from several sources, it does in fact quite well narrow the gap between the GF4 and 9700 in 3dmark.
Cheers,
-Shark
Definately need more testing to try and see I can get a configuration, registry or driver set to match Galilee's findings. Will keep trying to find the missing link, which admittedly my set of tests are half hearted at best on a "in use" PC which could be the culprit.
....
Seems to be the case with 3DMark and Aquamark at least. Not sure if this is always the case as more benchmarks would be needed.
Well, one could call 8x which is precisely 4x "just fine" since it indeed runs just fine.
As for backing up the claim, just flip back a couple pages in this thread with the custom level UT shots you guys provided. There *is* a performance difference between the identical 4x and 8x shots, just benchmark. The main bit of magic in the 40.xx drivers is that under some conditions, the nearly identical 4x and 8x dont show much of a performance hit (ala Aquamark) but do if you ramp up the textures or use a benchmark that isnt overly cpu/poly bound. (see the little charts as provided of 3dmark tests that show this trend quite plainly) you will see a difference in performance between 4x and 8x, so it cant simply be swept under the rug as "well, 8x just isnt working" since it will produce a different final score in, say, 3dmark?As has already been quoted on this very thread from several sources, it does in fact quite well narrow the gap between the GF4 and 9700 in 3dmark.
Cheers,
-Shark
Anyone know why there is a performance decrease in Giants?
I found performance went down by over 35%.
It is not all bad though as GP4 looks so much better now (my texture aliasing issue in another thread) and performance has increased by about 16% using 0x.
If someone else can benchmark Giants as well that would be great.
I found performance went down by over 35%.
It is not all bad though as GP4 looks so much better now (my texture aliasing issue in another thread) and performance has increased by about 16% using 0x.
If someone else can benchmark Giants as well that would be great.
LOL, Speed Tree Demo doesn't work now with the new Det. The first CG program supposenly. It gives all sorts of errors when I try to run it.
Why is QuakeIII all faded now as well? it looks gross besides being dark.
These are truely beta drivers, transpancy in W2K is interesting but not consistent either. I like the additional features and menu's but don't like the IQ of these drivers.
Why is QuakeIII all faded now as well? it looks gross besides being dark.
These are truely beta drivers, transpancy in W2K is interesting but not consistent either. I like the additional features and menu's but don't like the IQ of these drivers.
Who said anything about 2X and 4X? I've stated numerous times here the goal has been to improve max-AF (read 8x) performance and this has occurred.
There was quite a good discussion on this topic on this thread:
http://216.180.225.194/~beyond3d/forum/viewtopic.php?t=1343
Complete with screenshots (and even one delta shot between 4x and 8x) provided in that thread of previous driver revisions (as well as on the GF3), along with my AF test link w/ 29.42s, which once again no one has tried to answer the quiz... why? Because they are *freaking identical*, with the exception of a minimalized performance hit.
Cheers,
-Shark
QuakeIII is faded, colors are dull even after brightening it up, plus PowerStrip doesn't adjust the gamma like before . Serious Sam SE looks fine and it looks just to be a Quake III problem for me.
Transpancy doesn't work if I have a video stream going in Microsoft media player (bummer).
I will probably go back to the previous drivers, I really see no significant speed boost and IQ wise it looks worst for me.
. Serious Sam SE looks fine and it looks just to be a Quake III problem for me.Transpancy doesn't work if I have a video stream going in Microsoft media player (bummer).
I will probably go back to the previous drivers, I really see no significant speed boost and IQ wise it looks worst for me.
noko said:QuakeIII is faded, colors are dull even after brightening it up, plus PowerStrip doesn't adjust the gamma like before
Mine has the same problem. Hopefully this will be fixed in the upcoming Q3 point release.
But why won't these drivers allow more than a 75Hz refresh rate at 1,600 * 1,200 in 32 bit colour?
Even all my old resolutions above 1,900 * 1,200 have dissappaered (all the way up to 2,048 * 1,536).
Nothing I have thought of (haven't hacked the registery so far) will allow Windows 2000 to display these resolutions on a Sony 21" G500 monitor even on the desktop - let alone in a game.
Even all my old resolutions above 1,900 * 1,200 have dissappaered (all the way up to 2,048 * 1,536).
Nothing I have thought of (haven't hacked the registery so far) will allow Windows 2000 to display these resolutions on a Sony 21" G500 monitor even on the desktop - let alone in a game.
Looks like my QuakeIII issue of faded blue jeans I mean faded image is cleared up by using texture sharpening. I see, dull it for performance .
Anyways here are some shots from NOLF2 with the Fraps counter upper right telling you FPS at 1152x864x32 with following tests.
NOLF2 using 2xAA and No Anisotropic filtering (Tri-linear only)
NOLF2 using 2xAA and 2xAF
NOLF2 using 2xAA and 4xAF
NOLF2 using 2xAA and 8xAF
NOLF2 using 2xAA and 8xAF with texture sharpending
Notice what texture sharpening does in NOLF2, A hell of a difference, which in QuakeIII does take away the gross faded look, with a loss of performance. Nvidia is trading IQ for performance, but you can get it back if you check the texture sharpening box.
.Anyways here are some shots from NOLF2 with the Fraps counter upper right telling you FPS at 1152x864x32 with following tests.
NOLF2 using 2xAA and No Anisotropic filtering (Tri-linear only)
NOLF2 using 2xAA and 2xAF
NOLF2 using 2xAA and 4xAF
NOLF2 using 2xAA and 8xAF
NOLF2 using 2xAA and 8xAF with texture sharpending
Notice what texture sharpening does in NOLF2, A hell of a difference, which in QuakeIII does take away the gross faded look, with a loss of performance. Nvidia is trading IQ for performance, but you can get it back if you check the texture sharpening box
.I decided to check what would setting aniso level0 to stage 0 in RivaTuner would do to the framerates in conjuction with aniso 8.
And so I set aniso 8 in 40.41 driver panel, run RivaTuner, but for some strange reason, it shows me that 4x aniso is set...
So I decided to benchmark with 3dmark2k1se like that and got a score of 10272 (my score without Aniso in WinME on 30.82 was 10346!!).
Haven't checked the image quality yet, but according to ixbt, the IQ is basically the same (except Game 3 where there is blockiness and where setting the limit of level0 on stage1 removes it).
It's quite intresting actually, since the score I received with Aniso8 on 40.41 without limiting the aniso level on any stages was 8298.
Now to the weird thing: When I chose "force aniso 8" in Riva Tuner and then went to 40.41 Direct3D control panel, it showed me that the aniso level is set to 0x! Maybe RT doesn't work correctly with that new revision of drivers...
Gonna test today with these settings and see what happens...
Also, after setting the aniso level to 0x in the Direct3D control panel, I ran RT and it showed me that the aniso level is set to point samping! Definetly that bug Brian was talking about yesteday which will be fixed in the final version.
edit:
Alright, I tested with 3dmark2k1se with aniso level 8 set in Riva Tuner and with that setting, 0x set in Direct3D control panel and got a score of 9936.
And so I set aniso 8 in 40.41 driver panel, run RivaTuner, but for some strange reason, it shows me that 4x aniso is set...
So I decided to benchmark with 3dmark2k1se like that and got a score of 10272 (my score without Aniso in WinME on 30.82 was 10346!!).
Haven't checked the image quality yet, but according to ixbt, the IQ is basically the same (except Game 3 where there is blockiness and where setting the limit of level0 on stage1 removes it).
It's quite intresting actually, since the score I received with Aniso8 on 40.41 without limiting the aniso level on any stages was 8298.
Now to the weird thing: When I chose "force aniso 8" in Riva Tuner and then went to 40.41 Direct3D control panel, it showed me that the aniso level is set to 0x! Maybe RT doesn't work correctly with that new revision of drivers...
Gonna test today with these settings and see what happens...
Also, after setting the aniso level to 0x in the Direct3D control panel, I ran RT and it showed me that the aniso level is set to point samping! Definetly that bug Brian was talking about yesteday which will be fixed in the final version.
edit:
Alright, I tested with 3dmark2k1se with aniso level 8 set in Riva Tuner and with that setting, 0x set in Direct3D control panel and got a score of 9936.
Similar threads
