PDA

View Full Version : 65.52 vs 66.00 on a 6800 GT - VST, CS:S, Farcry, D3 etc


Pages : [1] 2

Dot50Cal
08-23-04, 09:40 PM
-Well I have some bad news. It appears there is either bug or an optimization which cannot be disabled in the 66.00 drivers. It makes the transition between mipmaps very hard thus resulting in better FPS. Ill be re-installing the 66.00's and making sure this is how it looks with all optimizations off. For reference I have included a .gif displaying this in action. You can see the 61.77's and the 65.52's do not have this optimization however the 66.00's do.

This optimization also fits the performance increase I have been seeing in my "High Quality" tests.

http://img13.exs.cx/img13/4952/opts.gif

-The Regulator issue also occured on the 61.77's when AA and AF were applied. I beleive the problem lies in the demo that was created. When AA and AF are applied it takes noticably longer for Jack to get outside. I beleive there are some things happening that are very intensive when Jack is about to enter the door. I will probably be discontinuing this demo as it produces inaccurate and false results.

-Graphs have been updated with the 61.77 drivers now.

-Doom 3 demo command added in D3 section

I've included a FAQ at the bottom for some questions I get sometimes. If you have any questions feel free to PM me.

Test system is as follows:
Pentium 4C 2.4 @ 3.0
1GB OCZ PC4000 (ddr500) 3-4-4-8
Evga Geforce 6800 GT @ 400/1100
Abit IC7-G Motherboard
SoundBlaster Audigy 2 ZS
Direct X 9.0C
Windows XP Service Pack 2
400 Watt Antec PSU

Far Cry
FarCry was being run in Shader Model 3.0 path as well as Geometry Instancing. Note: Geometry Instancing GREATLY impacts performance however it delivers the best IQ.

http://img69.exs.cx/img69/3849/fclow.jpg

http://img69.exs.cx/img69/4979/fchigh.jpg

With Far Cry I did notice a bug with the 55.52 drivers like I did in the last test I ran with them. While running the timedemo regulator the game froze when Jack was about to enter the door to the outside. It came back 30 seconds later however the FPS was impacted greatly by the delay. Therefore I am NOT including its number in the test.

Doom 3
Doom 3 was set to Ultra Quality. The command "timedemo demo1 usecache" was used (without quotes). This allows Doom 3 to cache the textures thus eliminating the need to run the demo two times in order to get accurate results.

http://img69.exs.cx/img69/5154/d3low.jpg

http://img69.exs.cx/img69/1963/d3high.jpg

With Doom 3 I did notice an IQ change from 65.52 and 66.00 in my High Quality test. See the bottom of the page for more information.


3DMark 03
Default settings were used however for low quality all optimizations were set to on and for high quality optimizations were set to off.

http://img69.exs.cx/img69/5854/3dmarklow.jpg

http://img69.exs.cx/img69/9902/3dmarkhigh.jpg

I also got the FPS scores from the first 4 game tests. They are as follows:

http://img69.exs.cx/img69/6108/3dmarkfpslow.jpg

http://img69.exs.cx/img69/8268/3dmarkfpshigh.jpg

Some IQ differences were also noted for 3DMark. See below for details.

Dot50Cal
08-23-04, 09:40 PM
Valve Stress Test
The Valve Stress Test is a test enviroment that simulates lots of the high end effects the Half-Life 2 engine can pull off. It is a VERY shader intensive map. Note: The VST and CS:S both seem to turn the reflections setting to world instead of all when exiting it. I counter acted this by setting it to all every time I restarted the game. Also it seem you will get inconsistent results if you do not exit the game completely and restart it in between tests.

http://img69.exs.cx/img69/3640/vstlow.jpg

http://img69.exs.cx/img69/8417/vsthigh.jpg


Counter-Strike Source
Since CS:S does not come with a built in timedemo I created one by using spectator mode to fly through the map (without players). I had initially tried to record a demo online in a match however the results were borked possibly due to the fact I am on 56k currently. What you are seeing here is the results from a flyby demo from de_dust.

http://img69.exs.cx/img69/5257/csslow.jpg

http://img69.exs.cx/img69/2448/csshigh.jpg

Image Quality Section:

The following images are .gif's from Doom 3 and 3DMark. I only noticed descrepencys in the images while running in my "High Quality" mode. Only Doom 3 and 3DMark exibited these quirks. I do not have enough knowlege to recoignize if any picture has "better" quality than the other however some do look noticiably sharper (Similar to ATI's?)

http://img43.exs.cx/img43/5107/1074.gif

http://img43.exs.cx/img43/2082/1075.gif

http://img43.exs.cx/img43/7121/1076.gif

http://img43.exs.cx/img43/4395/1077.gif

I hope you all enjoyed this, It took almost all my time today! :ORDER: Comments/Suggestions as always are appreciated.

FAQ:
"What is Low Quality"

Low Quality means there is no AA or AF being applied. Also all optimizations are set to ON and mip maps are set to NONE. All In-Game settings however are set to their maximum

"What is High Quality"

High Quality means 4xAA and 16xAF. All optimizations are set to OFF and mip maps are set to Trilinear. All In-Game settings are set to their maximum.

aAv7
08-23-04, 09:55 PM
Nice work! I appreciate your taking time out to do this. It seems based on the pics you posted above that at least in those instances-the 66.00's deliver lower i.q than the 65.52's, especially in that third shot. Do you agree?

I'm glad I read this. I was about to re-install the 66.00's after trying the 62.20's to see if rome total war performance would improve.... I'll stick with the 62.20' s now. I'm all for better frames...but not at the cost of image quality. Thankyou for the post.

Warlord
08-23-04, 10:14 PM
It seems based on the pics you posted above that at least in those instances-the 66.00's deliver lower i.q than the 65.52's, especially in that third shot. Do you agree?

Ahh I suggest you look again...I think you have it reversed....

anzak
08-23-04, 10:25 PM
Excellent job! :thumbsup: That is the best driver comparison I have ever read.

It seems based on the pics you posted above that at least in those instances-the 66.00's deliver lower i.q than the 65.52's, especially in that third shot. Do you agree?

The picture as a whole would look sharper. Thats just my opinion though.

The 66.00 driver seem to be a real winner. IQ enhancements and a performance boost at the same time, something that I have never seen from the catalyst driver team. I believe that the AF has been reworked this would explain the IQ enhancements along with improved performance when enabled.

FearMeAll
08-23-04, 10:31 PM
all I can say is wow. what a review.
hmm..I could never get the 66 drivers to work correctly anyways..always ran FSAA in slomotion it is really wierd.
But I'm very happy with 65.62s.
again..great review..must have taken a long time.

aAv7
08-23-04, 10:44 PM
I got the driver #'s mixed up..my fault. Thanks for not pulling a mezkal and insult me. Nice to know some people can be polite about pointing out an error, and understand as humans, we all make mistakes (except for the great mezkal himself of course.) =)

Dot50Cal
08-23-04, 10:51 PM
all I can say is wow. what a review.
hmm..I could never get the 66 drivers to work correctly anyways..always ran FSAA in slomotion it is really wierd.
But I'm very happy with 65.62s.
again..great review..must have taken a long time.

Hey! I noticed the same thing at one time in Doom 3. I was running 1600x1200, 4xAA 16xAF. I was getting half the FPS I should be. So I tested and it seems that my monitors refresh rate was the cause. Try setting it to default in the over-ride settings. That fixed my problem. Funny thing too cause I heard of this happening with V-Sync on and mine was forced to off. So anyone getting that I suggest setting your refresh rate to default.

Thanks for the comments everyone! Yeah the 66.00 do look sharper to me as well but in the one Doom 3 shot it seems to sharpen the right side wall a bit while blurring the area behind it a bit more, perhaps as someone mentioned NV reworked their AF settings. No complaints from me though, I could hardly notice it and it was the only area that was different. I guess one could liken it to ATI's situation with adaptive AF which when compared its not very clear which one is delivering the best IQ (Although the "artifacts"/sharpening are more apparent on the ATI cards than these 66.00's)

Also to be clear these differences were only seen when all optimizations were OFF. The optimization off images showed NO differences.

jAkUp
08-23-04, 11:32 PM
nice! I applaud your work :) And to me the 66.00 drivers seem to look better overall as far as quality goes

Blacklash
08-23-04, 11:43 PM
In the 66.00 driver anisotropic filtering now has both a mip and sample optimization alternative choice so it can be much like ATi's if you so desire. Same thing with the trilinear optimizations, there is a choice. Setting the quality slider to high quality defeats all three in one shot. This is where I game, I don't find the optimizations needed on my system and I always have a choice if I change my mind. Thank you nVidia.

And thank you, Dot50Cal, interesting read.

These days nVidia is definitely headed in the right direction.

Dot50Cal
08-23-04, 11:50 PM
Same here Malficar, I dont use optimizations either but since NV provides the option now (woo!) I think some people would be interested in seeing how it performs. I know quite a few people in the forums use it (even though it hardly raises framerates - check my 3dmark game tests)

The mip optimization was also in the 65.52's. Also wasnt the sample one in the 61.77's? I cant remember heh..Id love to get my hands on those 66.02's I read about.

Sazar
08-23-04, 11:55 PM
ep's b'day today so he's probably out getting drunk... thats why he hasn't posted in the 66.02 thread since his last one I spose :D

holmes
08-24-04, 12:12 AM
wow! looks good.. I gotta get my hands on the 66.00s..
:)

Electronic Punk
08-24-04, 03:00 AM
Optimisations make a difference when running 3dmark?

Dot50Cal
08-24-04, 03:13 AM
Yup, About 1-2 FPS average. Also im in the process now of updating the graphs which now include the 61.77's. IQ comparison with the 61.77's will be uploaded in about 30 minutes.

Skinner
08-24-04, 03:21 AM
Thanx, you have convinced me to install these ;) Nice job.

Dot50Cal
08-24-04, 04:26 AM
Id hold off on the install. Im trying to verify if there is an optimization or a bug (or user error) with these new drivers that you cannot turn off. See my first post in this thread for info.

Dot50Cal
08-24-04, 04:49 AM
Well I can now confirm that there is indeed a bug or optimization in these drivers which creates very harsh mip map transitions with the following settings even!
4xAA, 16xAF, All 3 Opts off, trilinear filtering.

Here is some information. Hopefully it will lead to an answer if this is a bug or an optimization.

-Renaming doom3.exe shows the same harsh transitions.

-This is noticed in other games but it is not as pronounced as it is in Doom 3 and 3D Mark.

-65.52 does not exibit this behavior nor does 61.77

If anyone has suggestions Id be glad to try em.

Arioch
08-24-04, 05:42 AM
My ATI X800XT had optimizations that were hardly noticeable in some games, but more noticeable in others. These optimizations seem to be more noticeable to some people than others and also more easy to spot out in a screenshot than in motion sometime. My guess is Nvidia is attempting to extract performance without impacting IQ and apparently they need more work on this, at least with these drivers. I guess that is why they are beta drivers after all.

Don't get me wrong, I'll take every optimization in the world if it does not impact IQ any. This applies to my old X800XT and current 6800 Ultra card. At least each new beta release shows some improvements in overall speed boosts and IQ.

Electronic Punk
08-24-04, 06:11 AM
Might have another go I guess. :o
How about 2001 will it make much of a difference there?

FearMeAll
08-24-04, 07:34 AM
Hey! I noticed the same thing at one time in Doom 3. I was running 1600x1200, 4xAA 16xAF. I was getting half the FPS I should be. So I tested and it seems that my monitors refresh rate was the cause. Try setting it to default in the over-ride settings. That fixed my problem. Funny thing too cause I heard of this happening with V-Sync on and mine was forced to off. So anyone getting that I suggest setting your refresh rate to default.

Thanks for the comments everyone! Yeah the 66.00 do look sharper to me as well but in the one Doom 3 shot it seems to sharpen the right side wall a bit while blurring the area behind it a bit more, perhaps as someone mentioned NV reworked their AF settings. No complaints from me though, I could hardly notice it and it was the only area that was different. I guess one could liken it to ATI's situation with adaptive AF which when compared its not very clear which one is delivering the best IQ (Although the "artifacts"/sharpening are more apparent on the ATI cards than these 66.00's)

Also to be clear these differences were only seen when all optimizations were OFF. The optimization off images showed NO differences.
I always keep vsync off and I still get honestly 1/4 the frames I should get with FSAA on. This is only with 66.00 drivers..the 65.62s and everything else do not have this problem. it's ok though..they're betas.
{edit}- oh it happened to you as well with vsync off hmmm...so setting it to default fixed it. interesting..I think I will stay with the 65s though and wait for the leaked 66.02.

OWA
08-24-04, 09:11 AM
Okay, I'm not following this thread very well. In the screenshots you showed the 66.00s look better to me (much sharper). But, this is a bad thing?

grimreefer
08-24-04, 11:14 AM
i have the feeling we r being watched....by onions!
http://discuss.futuremark.com/forum/showflat.pl?Cat=&Board=techdisplayadapters&Number=4271198&page=2&view=collapsed&sb=5&o=0&fpart=1

Dot50Cal
08-24-04, 03:44 PM
Okay, I'm not following this thread very well. In the screenshots you showed the 66.00s look better to me (much sharper). But, this is a bad thing?

Sharper does not mean better. Ive seen this in the past but I forget with what card (I believe it was an ATI card). Some textures/areas appear sharper but if you look close you can see that these sharper areas are extremely out of place. The areas surrounding it are not sharp but blurred a bit as they should be. Its most likely a bug but as you can see in the latest Doom 3 shot the mip maps are all messed up.

SpiderWoW
08-24-04, 04:45 PM
nice comparism - thx a lot

in my opinion the 66's pictures are "sharper" but more like using 24 bit for color depth instead of 32, so the color smoothing does not work properly and some colors are brighter than they should be in MY opinion. i will not install this version of the driver.