Subpar performance w/ 8800GTS

[Extelleron]

I just purchased an 8800GTS and got it yesterday, and the first thing I did was run 3D Mark 06. The score I got turned out to be LOWER than w/ my X1900XT (oc'd to 690/1565). I assumed it had something to do with drivers and maybe I hadn't done something right in switching over from ATI. So, I formatted and reinstalled XP today and immediately installed the TweaksRus 97.92 drivers and ran 3DMark again. The performance I got was pretty much exactly the same.


3DMark 06 scores-

Opteron 144 @ 2.4GHz, X1900XT @ 690/1565 - 4793 (March 25th, 2006)
Opteron 144 @ 2.4 GHz, 8800GTS @ 500/1600 - 4686 (Jan 17, 2007
Opteron 144 @ 2.4 GHz, 8800 GTS @ 650/1700 - 4873 (Jan 16, 2007)

Obviously something must be wrong. An 8800GTS should be way faster than the X1900XT even at stock, meanwhile here it can barely beat my X1900XT when overclocked @ 650/1700. Remember the 500/1600 test is with a fresh install of XP, so I don't see how it would be a problem with the drivers. All nVidia CP options are at their stock configurations.

Does anyone have any idea why I would be getting this kind of performance? I didn't spend $370 to get the same perfomance as my nearly one year old card.

BTW here is a link to my most recent test - http://service.futuremark.com/orb/proje...s.jsp?projectType=14&projectId=1027267

 

[LW07]

Did you use driver cleaner to clean off the ATI drivers after you uninstalled them?

 

[Extelleron]

Originally posted by: LW07
Did you use driver cleaner to clean off the ATI drivers after you uninstalled them?

Yes I did, but I formatted and reinstalled XP so that wouldn't matter anymore. The only drivers that have ever been installed as of now is 97.92. No other nVidia/ATI drivers have been installed.

 

[LW07]

Do your games perform any better?

 

[Extelleron]

The only game I've ever played on my GTS so far is CS:S, and that's not exactly a demanding game I'll reinstall it and try the video stress test, I know I got 110avg w/ my X1900XT 4x/16x.

Actually I really don't have any games that even stressed my X1900XT, and I didn't test for FPS either. I mainly bought the GTS for future games like Crysis/UT2007 and for running Vista when it comes out.

 

[BFG10K]

After you formatted did you install your chipset drivers?

 

[Extelleron]

Originally posted by: BFG10K
After you formatted did you install your chipset drivers?

Yes, that's the first thing I did. I have a DFI nForce 4 Ultra-D.

Update:

I ran the CS:S video stress test with 16xQ anti-aliasing, supersampling, 16xAF and most other CP settings at their highest levels. @ 1280x1024 res I got 121 FPS average. From memory I remember getting 110 FPS w/ my X1900XT at 4x adaptive AA/16xHQAF. I would think that 16xAA is alot more performance demanding, so perhaps my card just isn't up to par in 3DMark? I still can't see why it would be so low though.

Newb question: Which is better, multi-sampling or supersampling?

 

[ericeash]

i think the drivers still need to get sorted out. after a fresh install using 97.92 drivers on a fresh hdd, i got a 3dmark03 score(only test i can get results on since i don't have internet right now) of 27xxx on the rig in my sig at stock on the vid card. after installing a ton of programs and games, then oc'ing my 8800gts to 650/1950, my score as of today is 15xxx. not sure what i'm doing wrong, i have completely tested both my cpu/mem OCs as well as the GTS OC and both are perfectly stable. lots of other people reporting low scores as well.

 

[imported_dfifo]

Originally posted by: Extelleron
so perhaps my card just isn't up to par in 3DMark? I still can't see why it would be so low though.

We were discussing this at work today. If you look at benchmarks closely, alot of the top end cards show little difference running at resolutions of 1280x1024. Which is why most review sites are now listing 1600x1200 benchmarks as the standard.

It's the same reason if you look at older benchmarks you don't see reviewers compare cards at 800x600 resolution anymore because every benchmark would hit a wall at some point and cap the FPS. There's another bottleneck somewhere in the hardware.

This is probably what your experiencing with your system. If you can run a benchmark at 1600x1200 you'll probably see a more meaningful difference in the two cards, but I think in real world 1280x1024 gaming you're not going to see much difference until DX10 games start showing up.

 

[Cookie Monster]

Originally posted by: Extelleron

Originally posted by: BFG10K
After you formatted did you install your chipset drivers?

Yes, that's the first thing I did. I have a DFI nForce 4 Ultra-D.

Update:

I ran the CS:S video stress test with 16xQ anti-aliasing, supersampling, 16xAF and most other CP settings at their highest levels. @ 1280x1024 res I got 121 FPS average. From memory I remember getting 110 FPS w/ my X1900XT at 4x adaptive AA/16xHQAF. I would think that 16xAA is alot more performance demanding, so perhaps my card just isn't up to par in 3DMark? I still can't see why it would be so low though.

Newb question: Which is better, multi-sampling or supersampling?

Super sampling will always be better no matter what.

The reason 16xAA takes little performance hit (according to nVIDIA, 16xAA results in a 4xAA performance hit) because of the way they do AA. Its called CSAA (a type of MSAA). So basically nVIDIA has managed to allow you to enable higher levels of AA by using this "new" method. Note that it cannot work in some games due to the nature of the method.

Want me to go into it?

 

[Extelleron]

Originally posted by: Cookie Monster

Originally posted by: Extelleron

Originally posted by: BFG10K
After you formatted did you install your chipset drivers?

Yes, that's the first thing I did. I have a DFI nForce 4 Ultra-D.

Update:

I ran the CS:S video stress test with 16xQ anti-aliasing, supersampling, 16xAF and most other CP settings at their highest levels. @ 1280x1024 res I got 121 FPS average. From memory I remember getting 110 FPS w/ my X1900XT at 4x adaptive AA/16xHQAF. I would think that 16xAA is alot more performance demanding, so perhaps my card just isn't up to par in 3DMark? I still can't see why it would be so low though.

Newb question: Which is better, multi-sampling or supersampling?

Super sampling will always be better no matter what.

The reason 16xAA takes little performance hit (according to nVIDIA, 16xAA results in a 4xAA performance hit) because of the way they do AA. Its called CSAA (a type of MSAA). So basically nVIDIA has managed to allow you to enable higher levels of AA by using this "new" method. Note that it cannot work in some games due to the nature of the method.

Want me to go into it?

Seems to take a pretty big performance hit to me.

CS:S @ 4x/16x gives me 184 FPS (@ 650/1700)

CS:S @ 16xQAA/16xAF gives me 135 FPS (@ 650/1700)

 

[Cookie Monster]

Hmm heres a good explanation of how CSAA/cover sampling AA works.

Link

Basically 8xQ/16xQ takes much more bigger hit than 16xAA or 8xAA due to the difference in the number of samples taken. One thing to note is that 16xAA nor 8xAA really isnt giving you "real" 8xAA or 16xAA because of the nature of CSAA.

 

[TheRyuu]

Originally posted by: Extelleron

Originally posted by: Cookie Monster

Originally posted by: Extelleron

Originally posted by: BFG10K
After you formatted did you install your chipset drivers?

Yes, that's the first thing I did. I have a DFI nForce 4 Ultra-D.

Update:

I ran the CS:S video stress test with 16xQ anti-aliasing, supersampling, 16xAF and most other CP settings at their highest levels. @ 1280x1024 res I got 121 FPS average. From memory I remember getting 110 FPS w/ my X1900XT at 4x adaptive AA/16xHQAF. I would think that 16xAA is alot more performance demanding, so perhaps my card just isn't up to par in 3DMark? I still can't see why it would be so low though.

Newb question: Which is better, multi-sampling or supersampling?

Super sampling will always be better no matter what.

The reason 16xAA takes little performance hit (according to nVIDIA, 16xAA results in a 4xAA performance hit) because of the way they do AA. Its called CSAA (a type of MSAA). So basically nVIDIA has managed to allow you to enable higher levels of AA by using this "new" method. Note that it cannot work in some games due to the nature of the method.

Want me to go into it?

Seems to take a pretty big performance hit to me.

CS:S @ 4x/16x gives me 184 FPS (@ 650/1700)

CS:S @ 16xQAA/16xAF gives me 135 FPS (@ 650/1700)

Thats because your using 16xQ.
The regular 16xAA has the less of a performance hit.

This is how it all goes:
4xAA: 4xMSAA
8xAA: 4xMSAA+8xCSAA <-Looks similar to 8xMSAA but with performance hit of 4xMSAA
8xQ: 8xMSAA <-Large performance hit, bigger then 16x
16x: 4xMSAA+16xCSAA <-Looks better the 8xMSAA with slightly bigger performance hit of 4xMSAA
16xq: 8xMSAA+16xCSAA <-teh uber leet AA

So you can imagine why it's taking such a large performance hit. 8xMSAA is pretty harsh.
Regular 16x (4xMSAA+16xCSAA) seems to be a nice sweet spot sine the performance hit is only a little more then 8x (4xMSAA+8xCSAA) while having better IQ.

AF on the 8800GTS really doesn't make a hell of a difference in performance (according to the Rage3d article, IIRC).

And, AFAIK, you can't use super sampling with the 8800GTS (unless you can force it through nhancer...)

Also, you can use nhancer to change the settings which I think it's a hell of a lot easier to use nhancer to change both the global and profile settings. If I want to change monitor options though, I'm still stuck with the new nvidia CP (which I hate). Almost never go in there anymore. I do pretty much everything through Rivatuner and nhancer. And I've always liked Rivatuner for overclocking anyway (I think it's better then coolbits and loads better then ntune).

Good luck with your card.
And as a suggestion, I've always liked the XTreme-G Drivers for the quality and performance (linked version is the latest 97.92).

Just remember to use driver cleaner pro in safe mode.
And when you run 3dmark, use High Performance mode with all the opt. turned on with no AA/AF to get a better score.

 

[Extelleron]

Originally posted by: wizboy11

Originally posted by: Extelleron

Originally posted by: Cookie Monster

Originally posted by: Extelleron

Originally posted by: BFG10K
After you formatted did you install your chipset drivers?

Yes, that's the first thing I did. I have a DFI nForce 4 Ultra-D.

Update:

I ran the CS:S video stress test with 16xQ anti-aliasing, supersampling, 16xAF and most other CP settings at their highest levels. @ 1280x1024 res I got 121 FPS average. From memory I remember getting 110 FPS w/ my X1900XT at 4x adaptive AA/16xHQAF. I would think that 16xAA is alot more performance demanding, so perhaps my card just isn't up to par in 3DMark? I still can't see why it would be so low though.

Newb question: Which is better, multi-sampling or supersampling?

Super sampling will always be better no matter what.

The reason 16xAA takes little performance hit (according to nVIDIA, 16xAA results in a 4xAA performance hit) because of the way they do AA. Its called CSAA (a type of MSAA). So basically nVIDIA has managed to allow you to enable higher levels of AA by using this "new" method. Note that it cannot work in some games due to the nature of the method.

Want me to go into it?

Seems to take a pretty big performance hit to me.

CS:S @ 4x/16x gives me 184 FPS (@ 650/1700)

CS:S @ 16xQAA/16xAF gives me 135 FPS (@ 650/1700)

Thats because your using 16xQ.
The regular 16xAA has the less of a performance hit.

This is how it all goes:
4xAA: 4xMSAA
8xAA: 4xMSAA+8xCSAA <-Looks similar to 8xMSAA but with performance hit of 4xMSAA
8xQ: 8xMSAA <-Large performance hit, bigger then 16x
16x: 4xMSAA+16xCSAA <-Looks better the 8xMSAA with slightly bigger performance hit of 4xMSAA
16xq: 8xMSAA+16xCSAA <-teh uber leet AA

So you can imagine why it's taking such a large performance hit. 8xMSAA is pretty harsh.
Regular 16x (4xMSAA+16xCSAA) seems to be a nice sweet spot sine the performance hit is only a little more then 8x (4xMSAA+8xCSAA) while having better IQ.

AF on the 8800GTS really doesn't make a hell of a difference in performance (according to the Rage3d article, IIRC).

And, AFAIK, you can't use super sampling with the 8800GTS (unless you can force it through nhancer...)

Also, you can use nhancer to change the settings which I think it's a hell of a lot easier to use nhancer to change both the global and profile settings. If I want to change monitor options though, I'm still stuck with the new nvidia CP (which I hate). Almost never go in there anymore. I do pretty much everything through Rivatuner and nhancer. And I've always liked Rivatuner for overclocking anyway (I think it's better then coolbits and loads better then ntune).

Good luck with your card.
And as a suggestion, I've always liked the XTreme-G Drivers for the quality and performance (linked version is the latest 97.92).

Just remember to use driver cleaner pro in safe mode.
And when you run 3dmark, use High Performance mode with all the opt. turned on with no AA/AF to get a better score.

There's an option in the nVidia CP that allows you to choose Supersampling.

Anti-Aliasing - Transparency (Options are Off, Multisampling, or Supersampling)