nV News Forums

 
 

nV News Forums (http://www.nvnews.net/vbulletin/index.php)
-   NVIDIA Linux (http://www.nvnews.net/vbulletin/forumdisplay.php?f=14)
-   -   GTX470/GTX465 Terrible 2D Performance NVIDIA driver 275.09.07 (http://www.nvnews.net/vbulletin/showthread.php?t=164304)

AireTamStrm 07-11-11 12:12 PM

GTX470/GTX465 Terrible 2D Performance NVIDIA driver 275.09.07
 
1 Attachment(s)
I have had hell getting 2D to perform acceptably between a GTX470 and GTX465. The behavior is as many other threads describe: scrolling, resizing windows, or 2D-heavy operations just destroy the system. If I'm watching a video on one screen, on either GPU, scrolling a web browser on another screen (another GPU) will make the video studder and X will consume the CPU time of an entire core.

My BIOS by default configures large overlapping MTRR records. I have a bug with them on this, and in the mean time I've killed the uncachable and overlapping ranges by configuring them manually:

Before

reg00: base=0x000000000 ( 0MB), size= 4096MB, count=1: write-back
reg01: base=0x100000000 ( 4096MB), size= 1024MB, count=1: write-back
reg02: base=0x0c0000000 ( 3072MB), size= 1024MB, count=1: uncachable
reg03: base=0x0bf800000 ( 3064MB), size= 8MB, count=1: uncachable


After

cat /proc/mtrr
reg00: base=0x000000000 ( 0MB), size= 2048MB, count=1: write-back
reg01: base=0x100000000 ( 4096MB), size= 1024MB, count=1: write-back
reg02: base=0x080000000 ( 2048MB), size= 1024MB, count=1: write-back

I've heard both sides of the fence: MTRR matters, and MTRR doesn't matter. I have PAT support, so I figure the latter, but I also I figured hunting down that issue couldn't hurt. The MTRR settings take effect before the nvidia kernel module is loaded, but I'm not sure if its "too late" at that point. gtkperf benchmarks identically in both the "corrected" and default MTRR settings, suggesting that if MTRR is the problem, its too late.

As a side note: This is on a multiprocessor NUMA board (dual E5520 Xeons) with two IOHs. You will see that via the lspci output in the attached nvidia bug report. Also, I have a pair of GTX 260s collecting dust that I upgraded from, these /seem/ to not have the issue. I cannot recall 100% that they did or did not have the issue, so I will test them thoroughly when I get home.

I'm really lost as what to test next, any feedback at all would be extremely helpful.

AireTamStrm 07-19-11 01:39 PM

Re: GTX470/GTX465 Terrible 2D Performance NVIDIA driver 275.09.07
 
I'm still looking for assistance on this- is there any information I'm missing in my report?

drhirsch 07-22-11 05:23 AM

Re: GTX470/GTX465 Terrible 2D Performance NVIDIA driver 275.09.07
 
What kind of bug do you have with a overlapping MTRR configuration?

BTW, you can leave it to the kernel to reconfigure the MTRRs to have no overlap, at the cost of slightly more registers used with setting "MTRR cleanup" in the kernel .config or appending "enable_mtrr_cleanup" at the kernel parameter list.

AireTamStrm 07-27-11 11:00 AM

Re: GTX470/GTX465 Terrible 2D Performance NVIDIA driver 275.09.07
 
Quote:

Originally Posted by drhirsch (Post 2458552)
What kind of bug do you have with a overlapping MTRR configuration?

BTW, you can leave it to the kernel to reconfigure the MTRRs to have no overlap, at the cost of slightly more registers used with setting "MTRR cleanup" in the kernel .config or appending "enable_mtrr_cleanup" at the kernel parameter list.

I've tried both, and neither eliminate the overlap. The only way to kill the "uncacheable" entries is to manually specify the MTRRs myself.

2D, is unbearably slow. I had a chance to re-test my GTX 260, and it was (beyond) significantly faster. I could see traces of the issue, but it was so hard to reproduce with the 260 that I had to be messing with a web browser+some QT4 apps + etc in order to get it to jitter even -slightly-.

Could I get some help from NVIDIA on this? If there is anything in addition to the bug report needed, I need to know.

AireTamStrm 08-15-11 11:46 AM

Re: GTX470/GTX465 Terrible 2D Performance NVIDIA driver 275.09.07
 
*SOLVED* As an update to this issue:

I have to apologize to the community. I've seen countless threads like this on NVIDIA 2D performance and found that everyone either bent over and upgraded their distro or did some other -huge- change to their system that, while fixing the 2D performance issue, was not helpful to pinpoint the problem.

I'm afraid my case is similar. After being on OpenSuSE 11.3 for a long period of time, I got fed up with an unstable KDE installation (quite upgraded from distro stock), bad 2D performance, and clutter. I wasn't willing to fall back to a pair of GTX 260's to resolve the problem either.

I upgraded to the latest release of Arch Linux. On this new installation, my 465 and 470 have better 2D performance than the 260s did (which were quite acceptable) and there are no serious system hiccups (as there shouldn't be on an 8 Core / 16 Thread system). I have renewed hope for this system, which I was almost sure I would have to give up the massive CPU power for acceptable graphics performance.

Here's my best advice to you, who have stumbled onto this thread in hopes of a solution: You have no better excuse to try a new distro. Do a nice full backup, then run out and get something fresh. Read the documentation and give it a try. My Arch install was not without issue (the bootloader config was pointing to the wrong files - vmlinuz26 instead of vmlinux-linux, network did not setup correctly, etc), but using the doc I was able to overcome the issues and end up with a really nice, once-again solid desktop. I'm pleased with Arch, but I'm sure I'd have been just as pleased with Ubuntu, Debian, Fedora, etc....

If you're not satisfied with your current solution, try another one. Thats the beauty of the Linux desktop.


All times are GMT -5. The time now is 01:45 AM.

Powered by vBulletin® Version 3.7.1
Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.
Copyright 1998 - 2014, nV News.