PDA

View Full Version : How much does this texture setting affect performance?


Daneel Olivaw
11-05-04, 08:50 AM
Because it has a great picture quality impact. It makes anisotropic filtering almost uneeded imo.

rewt
11-05-04, 08:59 AM
It is supposed to force trilinear and anisotropic optimizations off. Performance should decrease by 15 - 20%.

*edit - but my tests prove otherwise, scroll down.

Daneel Olivaw
11-05-04, 09:12 AM
Good, because I'm tired of this brilinear filtering of Nvidia.

rewt
11-05-04, 09:12 AM
Ok I was totally wrong on that one. I just ran some benches with official Forceware 66.77. I got a higher score in Aquamark3 using High Quality than I did using Quality. This makes no sense... But hey I ain't complaining ;)

rewt
11-05-04, 12:32 PM
Doom3 is higher too

66.77 High Performance
49.9 FPS

66.77 High Quality
50.6 FPS !!

*both timedemos ran in cache mode

Daneel Olivaw
11-05-04, 12:36 PM
That is insane!? The image quality is unreasonably better, and the framerates improve?

superklye
11-05-04, 12:39 PM
hmmm...and this pickle you speak of only sweetens the deal.

rewt
11-05-04, 12:56 PM
That is insane!? The image quality is unreasonably better, and the framerates improve?

Yes, I don't understand it either. :retard: I though bilinear filtering and texture compression was supposed to make gameplay faster but for some reason it isn't. I wonder if this is some sort of bug in 66.77 Forceware?

I am going to run some tests on UT2004 next.

rewt
11-05-04, 01:05 PM
Same results.

br-colossus High Performance
18.104635 / 51.391270 / 102.658791 fps
Score = 51.470329

br-colossus High Quality
17.935452 / 51.420017 / 106.000237 fps
Score = 51.496925

High Quality yields better FPS no matter if AF is on or off! So the Aniso opts do not seem to be increasing performance in High Performance mode.

Daneel Olivaw
11-08-04, 09:17 AM
bump
So, what's up with this, any1 know?

CaptNKILL
11-08-04, 12:03 PM
Hmm... just an off the wall guess... but, do you think this could have something to do with the texture shimmering in some games? ie, an optimization that is supposed to be disabled with high-quality doesnt get disabled, so the performance stays almost the same no matter what the quality setting.

There really isnt anything to support this though, so dont start accusing them of cheating or something.

particleman
11-08-04, 01:00 PM
I find that the games that it has the biggest performance impact on is Unreal 2 engine games.

Daneel Olivaw
11-08-04, 04:36 PM
Mafia doesn't even look like the same game when using high quality texture filter. Btw, this shouldn't apply to opengl games (ie Doom3)

FearMeAll
11-08-04, 05:21 PM
ok I did this but it just ended up turning off anistropic filtering...but..it uh got rid of the shimmering.
BTW it also changed the control panel to say "high quality" too. What's strange is it still says 16x aniso as well.
I don't think this is the free quality people are hoping for.
BTW using 66.81 drivers.
ALso ran Mafia...and anistropic filtering was turned off..

Daneel Olivaw
11-08-04, 05:49 PM
ok I did this but it just ended up turning off anistropic filtering...but..it uh got rid of the shimmering.
BTW it also changed the control panel to say "high quality" too. What's strange is it still says 16x aniso as well.
I don't think this is the free quality people are hoping for.
BTW using 66.81 drivers.
ALso ran Mafia...and anistropic filtering was turned off..
and your conclusions? You do see a difference in Mafia no?

ChrisRay
11-08-04, 05:59 PM
I am going to have a look at this myself. I dont think texture filtering is key in Doom 3 since its mostly limited by other factors, a better test for testing Anisotropic Filtering performance would be a game that causes more texture swapping like unreal.

My Direct3d tests are still showing rather large discrepencies between the modes. High Quality Verses Quality(Opts Enabled) and High Performance verses Quality are all showing performance variations. I really do reccomend trying games which have more texture dependency since Doom 3 relies on EBM and Speculiar effects for its details. While Unreal/CoD will be more key to texture filtering.

rewt
11-08-04, 07:58 PM
Chris, I benched unreal tournament 2004. Very little difference in performance between High Performance and High Quality with AF on or off. In fact, High Quality looked 100 times better and was faster at most my benches.

And Doom3 I noticed faster at High Quality. As you know, High Quality turns off all the optimizations and intellisample texture compression that High Performance uses. So it makes absolutely no sense why it is faster. Maybe I just have magic 5900 ;)

particleman
11-08-04, 09:37 PM
It might be that you're CPU limited, especially if you are using a Athlon 1800+, make sure you bench with at least 4X AA and 8XAF and you will see a difference in UT2004 between the HQ and Q settings.

rewt
11-08-04, 09:46 PM
It might be that you're CPU limited, especially if you are using a Athlon 1800+, make sure you bench with at least 4X AA and 8XAF and you will see a difference in UT2004 between the HQ and Q settings.

But I am testing High Performance vs High Quality, in which there should be an even greater difference in performance!!

Unreal Tournament 2004, 4xAA / 8xAF
FX 5900NU @ 475/950
Forceware 66.93

High Performance
dm-rankin
18.110849 / 49.448792 / 95.248726 fps
Score = 49.521175

High Quality
dm-rankin
17.289495 / 49.521465 / 98.431267 fps
Score = 49.590660

High Quality is still faster!

ChrisRay
11-08-04, 09:50 PM
Chris, I benched unreal tournament 2004. Very little difference in performance between High Performance and High Quality with AF on or off. In fact, High Quality looked 100 times better and was faster at most my benches.

And Doom3 I noticed faster at High Quality. As you know, High Quality turns off all the optimizations and intellisample texture compression that High Performance uses. So it makes absolutely no sense why it is faster. Maybe I just have magic 5900 ;)


Thats interesting. I still cant recreate what you are seeing. My Performance analysis for AF still shows comparable results. Between High performance and High quality, (pretty large discrepency too) What resolution are you playing in? If so highten it, make yourself more fillrate bound. I dont have an NV35x to test with. So there could very well be something different about those chipsets. But the Nv4x line is still showing performance discrepencies. Btw try the "Ice Tomb" map. Its very bandwith/fillrate limited compared to others. Also Inferno is a good torture test too.

rewt
11-08-04, 09:55 PM
Resolution is 1280 x 1024 in UT2004. My monitor supports no higher. I will have to give those other maps a try.

Daneel Olivaw
11-17-04, 09:18 AM
I'm bumping my own thread since there was a question of HQ vs Q. Using the setting exposed in the first post, with a -0.5 mip bias, I've stopped using anisofiltering in DX games since the picture quality is so much better now.