crappy performance with 100.xx.xx series.

[linuxgameprogra]

Hello, on both my slackware setup, and a temporary installation of Ubuntu I used for testing, I get much crappier 3d performance with my new evga 680i motherboard and evga 8800 GTS 320MB.

If I revert to the 9xxx series, I get the performance I expect. With the recent console restoration fix in the new driver, (of which I am afflicted) I would really like to upgrade.

I have no idea what's causing this, and I'm really suprised I haven't heard anything about it.

Slackware 11.0, Xorg 7.1.1, Kernel 2.6.21-ck2

xorg.conf: http://pastebin.ca/584314
kernel config: http://pastebin.ca/584313
nvidia-settings.rc: http://pastebin.ca/584321
dmesg:http://pastebin.ca/584331

Like I said, this happened on stock ubuntu 7.04, so I would have to assume the usual amount of futzing I do with slackware hasn't messed anything up (such as the third party xorg packages im using), and ubuntu also uses later versions of xorg, and the kernel as well. Some kind of bios option messing me up? Driver support for 8800 series no good at this time? I doubt the latter, but I sold my 7900 GT, so I can't test that.

Please assist. Thank you.

[netllama]

Please generate and attach an nvidia-bug-report.log, and document how you're benchmarking performance, along with the expected & actual performance numbers.

[gbil]

Originally Posted by netllama

Please generate and attach an nvidia-bug-report.log, and document how you're benchmarking performance, along with the expected & actual performance numbers.

netllama this issue has been raised before by many people, including myself.

Personally I use World of Warcraft as a benchmark and I see about 5-10 less fps between 9xxx and 100.x series, 9xxx being faster.

To be honest in this case I don't see a need for a bug report since there is no error or any other warning, just less fps. Easily reproducible, hopefully you can confirm this within an hour using a game for benchmarking.

[netllama]

I've just looked through all your older posts, and I do not see any references to this issue.

If you expect further assistance, you'll need to provide all of the information that I've requested.

Thanks,
Lonni

[gbil]

Originally Posted by netllama

I've just looked through all your older posts, and I do not see any references to this issue.

If you expect further assistance, you'll need to provide all of the information that I've requested.

Thanks,
Lonni

Please see:

http://www.nvnews.net/vbulletin/show...59&postcount=7

and this is more general

http://www.nvnews.net/vbulletin/showthread.php?t=90102


I will provide you though with numerous numbers to see the fps difference and log files although I doubt you'll see something there.

[netllama]

You didn't provide an nvidia-bug-report.log in either of the posts you referenced. I have nothing further to say here until you provide the information that was requested.

Additionally, I'd like to know whether this problem is present anywhere other than WoW.

[gbil]

Originally Posted by netllama

You didn't provide an nvidia-bug-report.log in either of the posts you referenced. I have nothing further to say here until you provide the information that was requested.

Well I didn't get a response from nvidia in either post I referenced, but I guess you still have nothing further to say about that.

Additionally, I'd like to know whether this problem is present anywhere other than WoW.

I don't play any other game so yes WoW is the only place I can see the problem. I also run Xplane but with 9755 I get the Xid error+crash solved in 100.14.09+ drivers so I can't use that as benchmark.


Here are some numbers and bug reports - The numbers below are with the same kernel 2.6.21.4 vanilla, running the CPU's at max speed 2.13GHz no 'ondemand' governor set, display set at 1680x1050 at NVidia 8800GTS 320MB .

9755:

WoW fps: between 90 to 130 , around 115fps most of the time measured inside IF with no other people around.

100.14.11:

WoW fps: between 70 to 110, around 85 most of the time in the same place again with no people around.

This numbers of course, in both cases, are much smaller around 35fps in busy areas etc again with 9755 driver giving around 10fps more but I can't create a test-bed to report fps since the environment is busy places changes all the time (more - less people around).

I don't want to report glxgears numbers since this is no real benchmark but just for the record 9755 gives 4000 more fps.

I also installed pprace (old tuxracer I think) which reports about 250-280 for 100.14.11 driver and about 270-300 for the 9755 driver, again though I don't think this is a 'serious' benchmarking tool.

Hope you now have the numbers and logs and I'm willing to run other benchmarking programs if they are easy to obtain and run.

[netllama]

I'd be interested in the results you obtain from running ViewPerf:
http://www.spec.org/gpc/downloadindex.html

Also, is the performance deficit still present if you use Twinview instead of separate X screens?

[dpw2atox]

I also notice a performance hit using the nvidia 100.xx.xx series drivers using my 8600 GTS. In warcraft i previously on my 7600 GT had 0 issues and the game ran smooth except in heavily populated areas but still fairly well. Now the game slows up randomly so that its just plain choppy and after a few seconds, its fine again. its very odd. i am using a stock ubuntu 7.04 install, i removed the driver packages and installed the 100.14.11 drivers as per the instructions in the sticky post.

edit: also in wine and cedega, running the game in d3d mode causes my mouse cursor to flicker, where before with the 9x.xx drivers it wouldnt.

[gbil]

Originally Posted by netllama

I'd be interested in the results you obtain from running ViewPerf:
http://www.spec.org/gpc/downloadindex.html

Well I don't define 'easy to obtain' a 650MB download but anyway, here are the results:


100.14.11

Run All Summary
3dsmax-04 Weighted Geometric Mean = 15.14
catia-02 Weighted Geometric Mean = 14.92
ensight-03 Weighted Geometric Mean = 16.36
light-08 Weighted Geometric Mean = 12.97
maya-02 Weighted Geometric Mean = 30.34
proe-04 Weighted Geometric Mean = 11.44
sw-01 Weighted Geometric Mean = 18.85
ugnx-01 Weighted Geometric Mean = 4.539
tcvis-01 Weighted Geometric Mean = 5.105

9755

3dsmax-04 Weighted Geometric Mean = 11.91
catia-02 Weighted Geometric Mean = 13.98
ensight-03 Weighted Geometric Mean = 16.76
light-08 Weighted Geometric Mean = 10.97
maya-02 Weighted Geometric Mean = 30.47
proe-04 Weighted Geometric Mean = 10.71
sw-01 Weighted Geometric Mean = 18.13
ugnx-01 Weighted Geometric Mean = 4.321
tcvis-01 Weighted Geometric Mean = 5.413


Personally I can't understand why some are lower and some are higher, please shed a light here.

Also, is the performance deficit still present if you use Twinview instead of separate X screens?

Will test this but I'm not sure if this work the same as two separate X screens.


EDIT: I can't test with Twinview as the result is not the same, I have 2 different monitors and in separate X screens mode each monitor has it's own native resolution and bars in Gnome while in Twinview the bars are screwed up and so I can't work on this mode, so this is no use.

[LubosD]

I *have not* checked the performance myself, but I've heard some reports about performance loss too.

[linuxgameprogra]

Each test was run twice. Doom 3 1.3 timedemo demo1, and Counter Strike Source Video Test, on wine-0.9.37, with GLSL enabled.

9775
doom3: 134.0
css-test: 185.05

100.14.11
doom3: 120.7
css-test: 118.1

nvidia-bug-reports attached.

ps. what other fb driver can I use besides vesa in the kernel?