Go Back   nV News Forums > Linux Support Forums > NVIDIA Linux

Newegg Daily Deals

Reply
 
Thread Tools
Old 03-24-08, 09:41 AM   #1
jordimarch
Registered User
 
Join Date: Jan 2005
Posts: 5
Default MTRR: write-combining vs write-back performance

Greetings and hello to everybody in my first post,

I've been struggling with my MTRR settings until I finally got what I think is write-combining enabled. Well, x11perf/Xmark don't show any big improvement in its main areas, so I am not really sure if I did activate properly the MTRR range or not.

My system is a dual LGA771 Asus board, with currently only one Xeon 5335 quad processor, a 8600GT with 256MB RAM and 4GB RAM. My BIOS didn't know how to properly set up MTRR registers (even with "discrete" allocation under BIOS), so Linux kernel mapped out all available memory as write-back + some uncacheable overlapped (I guess) for the PCI devices that do need uncached access memory. lspci -v shows the following:

USB + some other stuff in the range of 0xb....
Nvidia @ 0xc0000000 with 256MB RAM.

So I tweaked /proc/mtrr to get this:

-------------------------------------
reg00: base=0x00000000 ( 0MB), size=2048MB: write-back, count=1
reg01: base=0x80000000 (2048MB), size= 512MB: write-back, count=1
reg02: base=0xa0000000 (2560MB), size= 256MB: write-back, count=1
reg03: base=0xb0000000 (2816MB), size= 256MB: uncachable, count=1
reg04: base=0xc0000000 (3072MB), size= 512MB: write-combining, count=1
reg05: base=0xe0000000 (3584MB), size= 512MB: write-back, count=1
reg06: base=0x100000000 (4096MB), size=4096MB: write-back, count=1
-------------------------------------

OK, this settings seem to work properly, graphics seem to be a little bit more fluent than with the old write-back, but an x11perf shows no improvement at all. For instance, I guess big regions like 500x500 opaque stippled rectangle should benefit from combining writes into a single burst, but I see no difference at all....

Maybe the poor results I get dealing with big images are because of the driver and not these MTRR settings... For example, comparing this 8600GT with an integrated Intel i965 card:
-osrect500 (opaque stippled rectangle 500x500): 2460000.0/sec vs 8400.0/sec
-shmputxy500 (ShmPutImage XY 500x500 square): 1730000.0/sec vs 6.7/sec (!!!)
the Intel i965 has got WC MTRR properly set up by BIOS, not hand-made like mine...

By the way, in many other areas, 2D is quite fast and 3D works pretty decent for my needs (Half Life 2 goes pretty well @ 1680x1050), but this comparison with the Intel card left me a little bit sad/upset/deceptioned/puzzled...

Any ideas? How can I be sure I'm properly using WC MTRR? X11 Shared memory seems to be a real bottleneck (see 6.7 vs 1730000) -> any idea how to get shared memory working properly?

Thank you very much. Any comment is really appreciated.
Jordi

--
http://www.devilmaster.org
jordimarch is offline   Reply With Quote
Old 03-24-08, 09:49 AM   #2
chunkey
#!/?*
 
Join Date: Oct 2004
Posts: 662
Default Re: MTRR: write-combining vs write-back performance

Quote:
Originally Posted by jordimarch

Any ideas? How can I be sure I'm properly using WC MTRR? X11 Shared memory seems to be a real bottleneck (see 6.7 vs 1730000) -> any idea how to get shared memory working properly?

Thank you very much. Any comment is really appreciated.
Jordi
Well, MTRR is obsolete/dead... it was replaced by the PAT long ago.
So that's not the bottleneck! You should take a look around, there are lots of threads that are dealing with the bad 2D performance of the 8XXX and 9XXX GF series.
chunkey is offline   Reply With Quote
Old 03-24-08, 10:01 AM   #3
jordimarch
Registered User
 
Join Date: Jan 2005
Posts: 5
Default Re: MTRR: write-combining vs write-back performance

Hi chunkey,

OK, that explains why I see no difference between WB/WC...

I'll repeat the search of posts without focusing in MTRR... Any ideas regarding Xshm though?

Thanks,
Jordi

--
http://www.devilmaster.org
jordimarch is offline   Reply With Quote
Old 03-24-08, 10:23 AM   #4
chunkey
#!/?*
 
Join Date: Oct 2004
Posts: 662
Default Re: MTRR: write-combining vs write-back performance

Quote:
Originally Posted by jordimarch
Hi chunkey,

OK, that explains why I see no difference between WB/WC...

I'll repeat the search of posts without focusing in MTRR... Any ideas regarding Xshm though?

Thanks,
Jordi
Well, the 2D Acceleration Framework was written by Intel... for Intel...

You can play with:
Code:
nvidia-settings -a InitialPixmapPlacement=2 -a GlyphCache=1
But it won't help much...
chunkey is offline   Reply With Quote
Old 03-24-08, 11:47 AM   #5
txf
Registered User
 
Join Date: Feb 2008
Posts: 272
Default Re: MTRR: write-combining vs write-back performance

Also bear in mind that :

Code:
nvidia-settings -a InitialPixmapPlacement=2 -a GlyphCache=1
needs the latest beta driver 171.06

altho I believe InitialPixmapPlacement exists in the 169 series too.
txf is offline   Reply With Quote
Old 03-25-08, 07:59 AM   #6
jordimarch
Registered User
 
Join Date: Jan 2005
Posts: 5
Default Re: MTRR: write-combining vs write-back performance

Hi,

yes, I tried with InitialPixmapPlacement=2/4 (169.09) and it indeed went much worse...

I'll try again when 171.06 leaves beta.

Thx,
Jordi

--
http://www.devilmaster.org
jordimarch is offline   Reply With Quote
Reply


Thread Tools

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Forum Jump


Similar Threads
Thread Thread Starter Forum Replies Last Post
My UT2003 Tweak Guide DXnfiniteFX Gaming Central 48 10-30-02 11:59 PM

All times are GMT -5. The time now is 03:00 AM.


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