PDA

View Full Version : 3DCenter Filtering Article


Pages : [1] 2

StealthHawk
10-11-03, 03:10 AM
http://www.3dcenter.org/artikel/detonator_52.14/index_e.php

Now in English! Yum. Good investigative journalism is great.

Hanners
10-11-03, 04:35 AM
I only skimmed through it, but it looks like a job well done. Interesting stuff. :)

Behemoth
10-11-03, 04:50 AM
We conclude at the moment that nVidia really improved shader performance with the 52.10/14 drivers.

It has yet to be proven whether nVidia can turn this theoretical advantage in more performance in real applications

anyone with 5800/5900 wanna check that the shader improvement of 52.10/14 can carry over to the real world applications such as ati's 9800 demos or some other dx9 demos? this is quite an improvement if it is real.

mrsabidji
10-11-03, 04:55 AM
Yup, really interesting indeed. They state that the good old UT2003 trilinear filter optimization is now applied to every applications on the FX cards (with 52.10 and 52.14 drivers). Hoppefully, this is only a bug and we'll see a real trilinear filter in the WHQL Dets 5x.xx release.
I had noticed a certain 'blurry-look' of the textures in the games I play, using Dets 52.13 myself (it reminded me of my old Voodoo3 3K, actually :D ) but in fact, I didn't have time to play much and I assumed it was because I had forgotten to set the AF level.
It seems that texture stage 1-7 are forced to use a 2xAF no matter what AF setting you're using :( . Hoppefully this will be corrected too...

mrsabidji

mrsabidji
10-11-03, 04:56 AM
Originally posted by Behemoth
anyone with 5800/5900 wanna check that the shader improvement of 52.10/14 can carry over to the real world applications such as ati's 9800 demos or some other dx9 demos? this is quite an improvement if it is real.

I'd be glad to help if you know any free/reliable shader benchmark I could download.

mrsabidji

edit : Didn't Skuzzy say he was programming one recently ? I think he said something like he only had to do some artwork then his benchmark-tool would be ready. I can't remember where or when I read that though...

MikeC
10-11-03, 05:02 AM
A statement from their conclusion after testing the 52.14 drivers :

"With this, nVidia offers its GeForceFX customers no correct trilinear filtering for the majority of currently available games!"

My first reaction was the pity I had for Anandtech...

I tested gameplay in UT2003 today after installing the 51.75 drivers and was shocked when I clearly saw the transition between mipmap levels from the ground textures. What makes it worse is that I was playing the game at full speed and even requested application level trilinear filtering.

"Official" drivers have not yet been delivered to reviewers by NVIDIA for the upcoming GeForce FX 5950 Ultra launch. I've had the 5950 Ultra for a week now and it's very frustrating not being able to take "official" benchmarks. If the drivers are delayed much longer, I won't be able to deliver half of what I had planned to do in our preview.

Fortunately I do have a contingency plan though :angel2:

volt
10-11-03, 05:11 AM
Originally posted by MikeC
Fortunately I do have a contingency plan though :angel2: [/B]

Want to enlighten us ? :p

Behemoth
10-11-03, 05:17 AM
Originally posted by mrsabidji
I'd be glad to help if you know any free/reliable shader benchmark I could download.

mrsabidji

you can try chimp demo, lava caves and gargoyle, download here (http://www.ati.com/developer/demos/r9800.html)
and maybe this (http://www.daionet.gr.jp/%7Emasa/rthdribl/index.html) as well.

run them on both 45.23 and 52.xx drivers then tell us the performance difference. many thanks. :)

mrsabidji
10-11-03, 05:22 AM
You're welcome ;) . See you later then...

mrsabidji sets mode +gone_testing mrsabidji

mrsabidji

Uttar
10-11-03, 06:28 AM
Originally posted by MikeC
Fortunately I do have a contingency plan though :angel2:

I doubt this is what you were planning, but if you want to punish nV...
Feel free to force your staff to quote all the "bad for nV" parts of my upcoming editorial.

I'm SURE nV would *love* to see every major news site focus on the negative part ;)


Uttar

mrsabidji
10-11-03, 06:58 AM
Hello Uttar :) !
Here are my bench results so far (using FRAPS) :

2003-10-11 13:40:17 - rthdribl (45.23 640x480 0xAA 4xAF)
Frames: 1243 - Time: 41110ms - Avg: 30.235 - Min: 27 - Max: 33

2003-10-11 13:12:21 - rthdribl (52.13 640x480 0xAA 4xAF)
Frames: 1534 - Time: 40750ms - Avg: 37.644 - Min: 35 - Max: 41

2003-10-11 13:41:45 - Gargoyle (45.23 1024x768 0xAA 4xAF)
Frames: 1004 - Time: 59953ms - Avg: 16.746 - Min: 15 - Max: 19

2003-10-11 13:19:00 - Gargoyle (52.13 1024x768 0xAA 4xAF)
Frames: 1004 - Time: 60031ms - Avg: 16.724 - Min: 16 - Max: 19

2003-10-11 13:43:34 - Caves (45.23 1024x768 0xAA 4xAF)
Frames: 12726 - Time: 331328ms - Avg: 38.409 - Min: 25 - Max: 56

2003-10-11 13:22:46 - Caves (52.13 1024x768 0xAA 4xAF)
Frames: 14416 - Time: 331328ms - Avg: 43.509 - Min: 29 - Max: 63

Chimp 1.1 would not run :
[AwFn.cpp] (line 2488): D3DAw Error: AwTexImage2D - Unable to create texture object

mrsabidji

PS : all tests run with F@H in background :D . The card is a MSI 5900TD128.

Behemoth
10-11-03, 07:18 AM
great. interesting results. can you do 0xAF comparisons, since 45.23's 4xAF is not 52.13's 4xAF, they are not apple to apple. and the so-called automatic shader optimization doesnt seem to help on Gargoyle...
and you can just use the rthdribl's benchmark function, its more accurate.

Hanners
10-11-03, 07:21 AM
Originally posted by mrsabidji
I'd be glad to help if you know any free/reliable shader benchmark I could download.

ShaderMark 2.0. I haven't actually tried it myself yet (shame on me :( ), but it should be pretty useful and the new version hasn't reached the clutches of nVidia's 'optimisation' team yet.

mrsabidji
10-11-03, 07:26 AM
Originally posted by Behemoth
great. interesting results. can you do 0xAF comparisons, since 45.23's 4xAF is not 52.13's 4xAF, they are not apple to apple. and the so-called automatic shader optimization doesnt seem to help on Gargoyle...
and you can just use the rthdribl's benchmark function, its more accurate.

At once , Sir. :D

Originally posted by Hanners
ShaderMark 2.0. I haven't actually tried it myself yet (shame on me :( ), but it should be pretty useful and the new version hasn't reached the clutches of nVidia's 'optimisation' team yet.

Okay, I'll give it a go, thanks.

mrsabidji

MikeC
10-11-03, 07:43 AM
Originally posted by Uttar
I doubt this is what you were planning, but if you want to punish nV...


It's not that I want to punish them, but NVIDIA has to be aware that it takes time to thouroughly review today's sophisticated graphics cards. And because I have a full time job, most of my work is done on the weekends. Two weekends have gone buy and I have nothing to show.

I think it would be awesome if web sites who have preview samples would collaborate on previews. Think about different web sites concentrating on one part of a preview - technical analysis, image quality comparisons, synthetic benchmarks, game benchmarks, cheat and optimization detection analysis, benchmarks against the competition, DVD playback, drivers, etc. Of couse I would do the gameplay tests :)

That reminds me that I need to put my contingency plan into action. So how are the latest Catalyst drivers performing ;)

Spliffstarr
10-11-03, 08:02 AM
Have you tried voicing your idea to the other webmasters of the other sites. Seems like a good idea to me, and i could imagine that it would cut down the amount of time needed when writing up the reviews .

Behemoth
10-11-03, 08:30 AM
Originally posted by mrsabidji
At once , Sir. :D

please take your time, gentleman ;)

Hanners
10-11-03, 08:57 AM
Originally posted by MikeC
I think it would be awesome if web sites who have preview samples would collaborate on previews. Think about different web sites concentrating on one part of a preview - technical analysis, image quality comparisons, synthetic benchmarks, game benchmarks, cheat and optimization detection analysis, benchmarks against the competition, DVD playback, drivers, etc. Of couse I would do the gameplay tests :)

Not a bad idea at all really, but you know how the old saying goes - "If you want a job doing properly, do it yourself". ;)

Of course, if you're too busy you could always send the card to someone else to review... :angel: ;)

Skuzzy
10-11-03, 09:06 AM
Originally posted by mrsabidji
edit : Didn't Skuzzy say he was programming one recently ? I think he said something like he only had to do some artwork then his benchmark-tool would be ready. I can't remember where or when I read that though...

You might have noticed I have been MIA this week from the forums. I did something stupid late last week (and I knew better, damn it) and got a care package in the mail Monday morning (this week).
Things are on hold, and that is about all I can say right now. Don't ask. I cannot talk about it.

mrsabidji
10-11-03, 10:05 AM
Whoops... I have kind of a strange problem : using 45.23 dets (reinstalled twice, safe mode etc, just in case) I can't get ATI demos to render correctly with App-mode AF (0xAF). Actually, only some of the effects/models are rendered(lava streams in Caves, the gargoyle and 'red jewel' in Gargoyle)... I'll keep my first 45.23 results for the moment and will try with the 52.13 and see what happens... Shadermark and rthdribl aren't affected by this 'bug'.

mrsabidji

edit : how do you activate rthdribl benchmark mode by the way ?

Behemoth
10-11-03, 12:24 PM
Originally posted by mrsabidji
Whoops... I have kind of a strange problem : using 45.23 dets (reinstalled twice, safe mode etc, just in case) I can't get ATI demos to render correctly with App-mode AF (0xAF). Actually, only some of the effects/models are rendered(lava streams in Caves, the gargoyle and 'red jewel' in Gargoyle)... I'll keep my first 45.23 results for the moment and will try with the 52.13 and see what happens... Shadermark and rthdribl aren't affected by this 'bug'.

mrsabidji

edit : how do you activate rthdribl benchmark mode by the way ?
oh sorry, i thought there was benchmark mode in rthdribl, there isnt.
as for the ati demos, its strange, my 5800(sold) run them all fine in any AF mode, even on chimp demo, with 45.23. i've also heard 5900 run them fine too with 45.23.
so you cant get them to run correctly now? even with your previous setting - 0xAA 4xAF? can you post some screenshots of the problems please? its gonna be interesting. :)

mrsabidji
10-11-03, 12:33 PM
It only occurs with Application/0xAF actually... that's weird. I'll post some screenshots as soon as possible and the other benchies results a bit later, no problem.

mrsabidji

edit : I downloaded the 45.23 directly from NV's ftp, FYI.
edit#2 : screenshots...

http://mrsabidji.00freehost.com/weird00.jpg http://mrsabidji.00freehost.com/weird01.jpg

fivefeet8
10-11-03, 02:20 PM
Originally posted by mrsabidji
It seems that texture stage 1-7 are forced to use a 2xAF no matter what AF setting you're using :( . Hoppefully this will be corrected too...

mrsabidji

Not if you set the AF option to "application". Then it does use 8xAF in UT2k3.

I posted these in another thread but:

http://f1.pg.briefcase.yahoo.com/bc/fivefeet8@sbcglobal.net/lst?.dir=/My+Documents/PC&.order=&.view=l&.src=bc&.done=http%3a//f1.pg.briefcase.yahoo.com/

The first shot is with the Dets 52.13 beta detonators. The second shot is taken from an article at 3dcenter.org in which they used a anti cheat script + modified drivers to force Trilinear Filtering. They were using the 44.65 dets btw. I downloaded the demo they had to take identical shots.

My shot has FSAA enabled. You can easily tell which is mines by looking at the bottom right of the shots.

mrsabidji
10-11-03, 02:25 PM
And finally, some benchmark results (long post) :

2003-10-11 20:04:00 - rthdribl (45.23 640x480 0xAA 0xAF)
Frames: 1805 - Time: 60141ms - Avg: 30.012 - Min: 28 - Max: 33

2003-10-11 21:22:15 - rthdribl (52.13 640x480 0xAA 0xAF)
Frames: 2351 - Time: 59672ms - Avg: 39.398 - Min: 37 - Max: 43

ShaderMark 2.0 :
video mode / device info
(1024x768) X8R8G8B8 (D24X8) vram used 56623104
HAL (pure hw vp): NVIDIA GeForce FX 5900 (Anti-Detect-Mode: off, gamma correction: DAC)
pixel shader version: 2_0
partial precision: off
number of render targets: 1

45.23 :
shader 2 (Per Pixel Diffuse Lighting): 89 fps 11.1862 mspf 447 rendered frames
shader 3 (Per Pixel Directional Light Shader (Phong)): 62 fps 16.1468 mspf 310 rendered frames
shader 5 (Per Pixel Spot Light Shader (Phong)): 50 fps 19.8747 mspf 252 rendered frames
shader 6 (Per Pixel Anisotropic Lighting): 63 fps 15.9407 mspf 314 rendered frames
shader 7 (Per Pixel Fresnel Reflections): 53 fps 18.7801 mspf 267 rendered frames
shader 9 (Per Pixel Car Surface Shader): 23 fps 42.9583 mspf 117 rendered frames
shader 10 (Per Pixel Environment Mapping): 115 fps 8.7003 mspf 575 rendered frames
shader 11 (Per Pixel Environment Bump Mapping): 99 fps 10.0583 mspf 498 rendered frames
shader 12 (Per Pixel Bump Mapping): 44 fps 22.7600 mspf 220 rendered frames
shader 13 (Per Pixel Shadowed Bump Mapping): 30 fps 33.4595 mspf 150 rendered frames
shader 14 (Per Pixel Veined Marble Shader): 33 fps 29.9319 mspf 168 rendered frames
shader 15 (Per Pixel Wood Shader): 46 fps 21.5059 mspf 233 rendered frames
shader 17 (Fur Shader With Anisotropic Lighting): 4 fps 232.3719 mspf 22 rendered frames
shader 18 (Per Pixel Refraction and Reflection Shader with Phong Lighting): 18 fps 54.1546 mspf 93 rendered frames

52.13 :
shader 2 (Per Pixel Diffuse Lighting): 118 fps 8.4857 mspf 590 rendered frames
shader 3 (Per Pixel Directional Light Shader (Phong)): 75 fps 13.2495 mspf 378 rendered frames
shader 5 (Per Pixel Spot Light Shader (Phong)): 65 fps 15.3389 mspf 326 rendered frames
shader 6 (Per Pixel Anisotropic Lighting): 73 fps 13.7174 mspf 365 rendered frames
shader 7 (Per Pixel Fresnel Reflections): 68 fps 14.6606 mspf 342 rendered frames
shader 9 (Per Pixel Car Surface Shader): 34 fps 29.3468 mspf 171 rendered frames
shader 10 (Per Pixel Environment Mapping): 144 fps 6.9423 mspf 721 rendered frames
shader 11 (Per Pixel Environment Bump Mapping): 116 fps 8.5843 mspf 583 rendered frames
shader 12 (Per Pixel Bump Mapping): 61 fps 16.4570 mspf 304 rendered frames
shader 13 (Per Pixel Shadowed Bump Mapping): 35 fps 28.2243 mspf 178 rendered frames
shader 14 (Per Pixel Veined Marble Shader): 47 fps 21.4634 mspf 233 rendered frames
shader 15 (Per Pixel Wood Shader): 46 fps 21.8437 mspf 229 rendered frames
shader 16 (Per Pixel Tile Shader): 36 fps 27.9297 mspf 180 rendered frames
shader 17 (Fur Shader With Anisotropic Lighting): 5 fps 185.1371 mspf 28 rendered frames
shader 18 (Per Pixel Refraction and Reflection Shader with Phong Lighting): 26 fps 38.1248 mspf 132 rendered frames

mrsabidji

edit : 0xAA and 0xAF used for SM2.0 too

mrsabidji
10-11-03, 02:27 PM
Halo timedemos (0xAA 0xAF):

Date / Time: 11/10/2003 17:37:13 (159937ms)
1700MHz, 512MB, 128M nVidia GeForceFX 5900 (DeviceID=0x0331) Driver=6.14.10.4523 Shader=2.0
F:\Halo\Halo.exe -vidmode 1024,768,85 -timedemo (Version=1.0.1.580)
Frames=4700
Total Time=190.51s
Average frame rate=24.67fps
Below 5fps= 9% (time) 0% (frames) (18.380s spent in 14 frames)
Below 10fps= 9% (time) 0% (frames)
Below 15fps= 12% (time) 1% (frames)
Below 20fps= 26% (time) 12% (frames)
Below 25fps= 55% (time) 38% (frames)
Below 30fps= 71% (time) 55% (frames)
Below 40fps= 92% (time) 84% (frames)
Below 50fps= 97% (time) 94% (frames)
Below 60fps= 99% (time) 97% (frames)
Memory used Max=170MB, Min=137MB, Ave=158MB

Date / Time: 11/10/2003 21:33:52 (4445468ms)
1700MHz, 512MB, 128M nVidia GeForceFX 5900 (DeviceID=0x0331) Driver=6.14.10.5213 Shader=2.0
F:\Halo\Halo.exe -vidmode 1024,768,85 -timedemo (Version=1.0.1.580)
Frames=4700
Total Time=171.64s
Average frame rate=27.38fps
Below 5fps= 12% (time) 0% (frames) (20.995s spent in 16 frames)
Below 10fps= 12% (time) 0% (frames)
Below 15fps= 14% (time) 1% (frames)
Below 20fps= 17% (time) 3% (frames)
Below 25fps= 33% (time) 17% (frames)
Below 30fps= 60% (time) 43% (frames)
Below 40fps= 84% (time) 73% (frames)
Below 50fps= 96% (time) 92% (frames)
Below 60fps= 99% (time) 97% (frames)
Memory used Max=169MB, Min=137MB, Ave=157MB

mrsabidji