680 GTX: Terrible DX9 Performance!
  1 / 2    
Hello All,

A few days ago I purchased a 680 GTX to use in my triple screen racing simulator rig. I did a clean install of Win7, installed all drivers and the iRacing motorsports simulator (www.iracing.com), which is the biggest racing simulator title in the world. iRacing uses DirectX 9.
I went out on the track and was shocked by the incredibly low FPS I got with the 680 GTX!

I started to look for problems with my rig, but after a while I discovered that I was not alone getting bad performance with the 680 GTX. Another iRacing community member has made an extensive comparison between the 680 GTX and the ATI 7970, and the 7970 is on average almost 70% (!) faster than the 680 GTX /omg.gif' class='bbc_emoticon' alt=':omg:' /> Here are the results of his test with 4x and 8x AA:

[img]http://i122.photobucket.com/albums/o251/px_seven/680_GTX_vs_7970_iRacing.png[/img]
Here is his rig specs:

CPU: Core i5 2500k @4.7GHz
RAM: 16GB DDR3 1600
MB: ASRock Extreme4 Gen3 Z68
Displays: 3x Asus 23" 1080p

I have a similar setup (i7 2600K based) and get very similar results as above, so there is nothing wrong with his rig. Also other iRacing community members have duplicated the results.

Obviously there is something terribly unoptimized in the Nvidia 680 GTX drivers for DX9, and it is not only in iRacing. He also did tests using the unigine benchmark, where the 680 GTX also performed very bad in DX9. Link to results: [url="https://docs.google.com/spreadsheet/ccc?key=0Al_P5Z01QUt0dFpqbjZfQ2ozemNsYlpsVUp0RWh3VXc#gid=0"]unigine results[/url]

in iRacing, the 680 GTX is even ~25% slower than my old ATI 6970 /omg.gif' class='bbc_emoticon' alt=':omg:' /> I thought I made a performance upgrade going from the 6970 to the 680 GTX, but instead I made a multi $$$ downgrade...

Nvidia, you [b]really[/b] need to fix the performance of your drivers all across the DX9 API, not only so that it is optimized for the big block buster DX9 titles like Skyrim etc.

I hope and expect to see Nvidia staff respond to this post, as the current state of DX9 performance is clearly unacceptable for such a high end, expensive card.

Thanks,
Hello All,



A few days ago I purchased a 680 GTX to use in my triple screen racing simulator rig. I did a clean install of Win7, installed all drivers and the iRacing motorsports simulator (www.iracing.com), which is the biggest racing simulator title in the world. iRacing uses DirectX 9.

I went out on the track and was shocked by the incredibly low FPS I got with the 680 GTX!



I started to look for problems with my rig, but after a while I discovered that I was not alone getting bad performance with the 680 GTX. Another iRacing community member has made an extensive comparison between the 680 GTX and the ATI 7970, and the 7970 is on average almost 70% (!) faster than the 680 GTX /omg.gif' class='bbc_emoticon' alt=':omg:' /> Here are the results of his test with 4x and 8x AA:



Image

Here is his rig specs:



CPU: Core i5 2500k @4.7GHz

RAM: 16GB DDR3 1600

MB: ASRock Extreme4 Gen3 Z68

Displays: 3x Asus 23" 1080p



I have a similar setup (i7 2600K based) and get very similar results as above, so there is nothing wrong with his rig. Also other iRacing community members have duplicated the results.



Obviously there is something terribly unoptimized in the Nvidia 680 GTX drivers for DX9, and it is not only in iRacing. He also did tests using the unigine benchmark, where the 680 GTX also performed very bad in DX9. Link to results: unigine results



in iRacing, the 680 GTX is even ~25% slower than my old ATI 6970 /omg.gif' class='bbc_emoticon' alt=':omg:' /> I thought I made a performance upgrade going from the 6970 to the 680 GTX, but instead I made a multi $$$ downgrade...



Nvidia, you really need to fix the performance of your drivers all across the DX9 API, not only so that it is optimized for the big block buster DX9 titles like Skyrim etc.



I hope and expect to see Nvidia staff respond to this post, as the current state of DX9 performance is clearly unacceptable for such a high end, expensive card.



Thanks,

#1
Posted 05/12/2012 07:03 PM   
D3D9 requires more cpu optimisations in the driver than D3D11 does, just to keep in mind.
D3D9 requires more cpu optimisations in the driver than D3D11 does, just to keep in mind.



In Memory of Chris "ChrisRay" Arthington, 1982-2010

CPU:Intel i7 920 @ 3.8(D0), Mainboard:Asus Rampage II Gene, Memory:12GB Corsair Vengeance 1600
Video:EVGA Geforce GTX 680+ 4GB, Sound:Creative XFI Titanium Fatal1ty Pro, Monitor:BenQ G2400WD
HDD:500GB Spinpoint F3, 1TB WD Black, 2TB WD Red, 1TB WD Black
Case:NZXT Guardian 921RB, PSU:Corsair 620HX, OS:Windows 7 SP1

#2
Posted 05/12/2012 10:44 PM   
Are you seeing low gpu usage according to Afterburner or Precession while seeing these low fps? If so people are seeing the same issue in other titles as well with GTX680.
Are you seeing low gpu usage according to Afterburner or Precession while seeing these low fps? If so people are seeing the same issue in other titles as well with GTX680.

#3
Posted 05/13/2012 12:32 AM   
[quote name='BababooeyHTJ' date='12 May 2012 - 05:32 PM' timestamp='1336869124' post='1407564']
Are you seeing low gpu usage according to Afterburner or Precession while seeing these low fps? If so people are seeing the same issue in other titles as well with GTX680.
[/quote]

Hi Folks,
The OP sent me the link to this discussion and since I'm the one that ran the 680 vs 7970 iRacing tests he referred to, I figured I'd chime in here too. To answer the question above, the GPU usage was approx 97-99%. That was pretty consistent across all the tests. It was interesting that I never saw 100% but I think that was likely just a quirk with how GPU-z was reporting it. For many of the tests I saw lengthy continuous 99% usage. I can link you guys to more test results if you like but they tell the same story. I did see an improvement from 300.65 to 301.10 and again when I went to 301.24 so hopefully this is a known issue nvidia is working on. The results quoted above were achieved with 301.24.
[quote name='BababooeyHTJ' date='12 May 2012 - 05:32 PM' timestamp='1336869124' post='1407564']

Are you seeing low gpu usage according to Afterburner or Precession while seeing these low fps? If so people are seeing the same issue in other titles as well with GTX680.





Hi Folks,

The OP sent me the link to this discussion and since I'm the one that ran the 680 vs 7970 iRacing tests he referred to, I figured I'd chime in here too. To answer the question above, the GPU usage was approx 97-99%. That was pretty consistent across all the tests. It was interesting that I never saw 100% but I think that was likely just a quirk with how GPU-z was reporting it. For many of the tests I saw lengthy continuous 99% usage. I can link you guys to more test results if you like but they tell the same story. I did see an improvement from 300.65 to 301.10 and again when I went to 301.24 so hopefully this is a known issue nvidia is working on. The results quoted above were achieved with 301.24.

#4
Posted 05/13/2012 01:37 AM   
if the GPU usage was that high, then you have not got a performance issue with the driver

what you are seeing is likely a change in the actual hardware design having an impact on performance. Quite likely, both Unigine and that racing game use an unusual technique that the new hardware doesn't execute as quickly as it did previously.

These minute regressions in D3D9 have been expected to start occurring for some time, as the hardware moves further into Pure D3D11 designs.

Of course, it could also mean that under d3d11, unigine uses better optimised texture and HDR compression formats that improve processing performance while maintaining similar visual quality
if the GPU usage was that high, then you have not got a performance issue with the driver



what you are seeing is likely a change in the actual hardware design having an impact on performance. Quite likely, both Unigine and that racing game use an unusual technique that the new hardware doesn't execute as quickly as it did previously.



These minute regressions in D3D9 have been expected to start occurring for some time, as the hardware moves further into Pure D3D11 designs.



Of course, it could also mean that under d3d11, unigine uses better optimised texture and HDR compression formats that improve processing performance while maintaining similar visual quality



In Memory of Chris "ChrisRay" Arthington, 1982-2010

CPU:Intel i7 920 @ 3.8(D0), Mainboard:Asus Rampage II Gene, Memory:12GB Corsair Vengeance 1600
Video:EVGA Geforce GTX 680+ 4GB, Sound:Creative XFI Titanium Fatal1ty Pro, Monitor:BenQ G2400WD
HDD:500GB Spinpoint F3, 1TB WD Black, 2TB WD Red, 1TB WD Black
Case:NZXT Guardian 921RB, PSU:Corsair 620HX, OS:Windows 7 SP1

#5
Posted 05/13/2012 02:02 AM   
For further reference I figured I would add the complete test results:

iRacing: https://docs.google.com/spreadsheet/ccc?key=0Al_P5Z01QUt0dDJwbXcyWEpaUHZOTDZ1ZEQtNXFqbkE#gid=0

Unigine: https://docs.google.com/spreadsheet/ccc?key=0Al_P5Z01QUt0dFpqbjZfQ2ozemNsYlpsVUp0RWh3VXc#gid=0
For further reference I figured I would add the complete test results:



iRacing: https://docs.google.com/spreadsheet/ccc?key=0Al_P5Z01QUt0dDJwbXcyWEpaUHZOTDZ1ZEQtNXFqbkE#gid=0



Unigine: https://docs.google.com/spreadsheet/ccc?key=0Al_P5Z01QUt0dFpqbjZfQ2ozemNsYlpsVUp0RWh3VXc#gid=0

#6
Posted 05/13/2012 05:34 AM   
i have to add to that. just did a few tests with heaven benchmark, and i get 15 fps less with dx9 than dx11 with same settings set

Render:
direct3d9
Mode:
1920x1080 4xAA fullscreen
Shaders:
high
Textures:
high
Filter:
trilinear
Anisotropy:
4x
Occlusion:
enabled
Refraction:
enabled
Volumetric:
enabled
Tessellation: disabled


Heaven Benchmark v3.0 Basic

FPS:
77.4
Scores:
1950
Min FPS:
43.8
Max FPS:
197.4


and then dx11 same settings
Settings

Render:
direct3d11
Mode:
1920x1080 4xAA fullscreen
Shaders:
high
Textures:
high
Filter:
trilinear
Anisotropy:
4x
Occlusion:
enabled
Refraction:
enabled
Volumetric:
enabled
Tessellation: disabled

Heaven Benchmark v3.0 Basic

FPS:
93.0
Scores:
2342
Min FPS:
27.2
Max FPS:
189.6
i have to add to that. just did a few tests with heaven benchmark, and i get 15 fps less with dx9 than dx11 with same settings set



Render:

direct3d9

Mode:

1920x1080 4xAA fullscreen

Shaders:

high

Textures:

high

Filter:

trilinear

Anisotropy:

4x

Occlusion:

enabled

Refraction:

enabled

Volumetric:

enabled

Tessellation: disabled





Heaven Benchmark v3.0 Basic



FPS:

77.4

Scores:

1950

Min FPS:

43.8

Max FPS:

197.4





and then dx11 same settings

Settings



Render:

direct3d11

Mode:

1920x1080 4xAA fullscreen

Shaders:

high

Textures:

high

Filter:

trilinear

Anisotropy:

4x

Occlusion:

enabled

Refraction:

enabled

Volumetric:

enabled

Tessellation: disabled



Heaven Benchmark v3.0 Basic



FPS:

93.0

Scores:

2342

Min FPS:

27.2

Max FPS:

189.6

#7
Posted 05/13/2012 05:43 PM   
[quote name='Sora' date='12 May 2012 - 10:44 PM' timestamp='1336862679' post='1407545']
D3D9 requires more cpu optimisations in the driver than D3D11 does, just to keep in mind.
[/quote]

that doesnt matter when an ati card performs the test on same cpu 20 fps faster

http://forums.nvidia.com/index.php?showtopic=228873 another person did some tests

http://forums.anandtech.com/showthread.php?t=2243854

here is another
http://forums.nvidia.com/index.php?showtopic=227564
[quote name='Sora' date='12 May 2012 - 10:44 PM' timestamp='1336862679' post='1407545']

D3D9 requires more cpu optimisations in the driver than D3D11 does, just to keep in mind.





that doesnt matter when an ati card performs the test on same cpu 20 fps faster



http://forums.nvidia.com/index.php?showtopic=228873
another person did some tests



http://forums.anandtech.com/showthread.php?t=2243854



here is another

http://forums.nvidia.com/index.php?showtopic=227564

#8
Posted 05/13/2012 05:48 PM   
Issues with CPU optimizations does come up every now and again, and ATI/AMD usually has an advantage.
Issues with CPU optimizations does come up every now and again, and ATI/AMD usually has an advantage.

"This is your code. These are also your bugs. Really. Yes, the API runtime and the
driver have bugs, but this is not one of them. Now go fix it already." -fgiesen

#9
Posted 05/13/2012 05:57 PM   
diablo 3 using dx9, lets hope nvidia comes out with a good driver for this game
diablo 3 using dx9, lets hope nvidia comes out with a good driver for this game

#10
Posted 05/14/2012 06:21 PM   
Filed a bug. Thanks. Quick couple of questions:

1) In the benchmarks, have you tried running the game with "Power Management Mode" in the NVIDIA Control Panel -> Manage 3D Settings set to "Performance"?

2) Does this occur in non-overclocked GTX 680 cards as well?
Filed a bug. Thanks. Quick couple of questions:



1) In the benchmarks, have you tried running the game with "Power Management Mode" in the NVIDIA Control Panel -> Manage 3D Settings set to "Performance"?



2) Does this occur in non-overclocked GTX 680 cards as well?

Please send me a PM if I fail to keep up on replying in any specific thread or leave a driver feedback: Driver Feedback

#11
Posted 05/14/2012 09:05 PM   
Hi Manuel,
All the tests I performed (and referenced above in the google docs links) were run in Performance mode. As for the non-overclocked cards, several fellow iRacing members ran some comparison tests on their cards - all of which were reference and at stock clocks afaik and the results were comparable. i.e. They scaled according to the clock settings. If you need any further info on my set of tests please let me know.

Brendan.
Hi Manuel,

All the tests I performed (and referenced above in the google docs links) were run in Performance mode. As for the non-overclocked cards, several fellow iRacing members ran some comparison tests on their cards - all of which were reference and at stock clocks afaik and the results were comparable. i.e. They scaled according to the clock settings. If you need any further info on my set of tests please let me know.



Brendan.

#12
Posted 05/14/2012 09:37 PM   
[quote name='ManuelG' date='14 May 2012 - 05:05 PM' timestamp='1337029501' post='1408268']
Filed a bug. Thanks. Quick couple of questions:

1) In the benchmarks, have you tried running the game with "Power Management Mode" in the NVIDIA Control Panel -> Manage 3D Settings set to "Performance"?

2) Does this occur in non-overclocked GTX 680 cards as well?
[/quote]

Add GTA4 and Serious Sam HD to the list of DX9 games with bad performance and low gpu usage with GTX680. There are quite a few reports of it in the GTX600 section that nobody reads apparently. [url="http://forums.nvidia.com/index.php?showtopic=227564"]There are a few other games mentioned in this thread.[/url] Also, most games take a much larger hit from 3d vision than they should. For example Skyrim runs like garbage.
[quote name='ManuelG' date='14 May 2012 - 05:05 PM' timestamp='1337029501' post='1408268']

Filed a bug. Thanks. Quick couple of questions:



1) In the benchmarks, have you tried running the game with "Power Management Mode" in the NVIDIA Control Panel -> Manage 3D Settings set to "Performance"?



2) Does this occur in non-overclocked GTX 680 cards as well?





Add GTA4 and Serious Sam HD to the list of DX9 games with bad performance and low gpu usage with GTX680. There are quite a few reports of it in the GTX600 section that nobody reads apparently. There are a few other games mentioned in this thread. Also, most games take a much larger hit from 3d vision than they should. For example Skyrim runs like garbage.

#13
Posted 05/14/2012 10:26 PM   
[quote name='brendanh' date='14 May 2012 - 02:37 PM' timestamp='1337031477' post='1408283']
Hi Manuel,
All the tests I performed (and referenced above in the google docs links) were run in Performance mode. As for the non-overclocked cards, several fellow iRacing members ran some comparison tests on their cards - all of which were reference and at stock clocks afaik and the results were comparable. i.e. They scaled according to the clock settings. If you need any further info on my set of tests please let me know.

Brendan.
[/quote]

Thanks for the update.
[quote name='brendanh' date='14 May 2012 - 02:37 PM' timestamp='1337031477' post='1408283']

Hi Manuel,

All the tests I performed (and referenced above in the google docs links) were run in Performance mode. As for the non-overclocked cards, several fellow iRacing members ran some comparison tests on their cards - all of which were reference and at stock clocks afaik and the results were comparable. i.e. They scaled according to the clock settings. If you need any further info on my set of tests please let me know.



Brendan.





Thanks for the update.

Please send me a PM if I fail to keep up on replying in any specific thread or leave a driver feedback: Driver Feedback

#14
Posted 05/14/2012 10:27 PM   
Why was this issue not picked up in all the 680 reviews recently? I know reviewers tend to focus more on DX11 titles, but there were [i]some[/i] DX9 titles, and a lot of reviews. Does it not apply as much to AAA titles?
Why was this issue not picked up in all the 680 reviews recently? I know reviewers tend to focus more on DX11 titles, but there were some DX9 titles, and a lot of reviews. Does it not apply as much to AAA titles?

i7 950 @4.0Ghz
MSI GTX 680 Lightning
Velociraptor 600GB
6GB RAM
GA-EX58-EXTREME
Corsair TX750w
___________________

Give it up for the President of America!! Dwayne Elizondo Mountain Dew Herbert Camacho!!
5-TIME ULTIMATE SMACKDOWN CHAMPION, PORN SUPERSTAR, AND PRESIDENT OF THE UNITED STATES

#15
Posted 05/15/2012 10:00 AM   
  1 / 2    
Scroll To Top