[ale] /proc/mtrr anyone?

[Eric Z. Ayers]

You can find a discussion of this on the SMP mailing list, and the
fellow who wrote the patch has a web site (maybe it's referenced in
the source?).  Basically, it believe it's nutsing with some
timers/features in the PCI bus to give you this speedup for bitblt
operations.    

And actually, I think you're just getting your (relative) speed back.  You
probably didn't notice BAD performance, but it was about the same
speed as a pentium, and should have been much faster.

-Eric.

Mike Kachline writes:
 > 
 > 	Today, I decided to give the 2.1.x series a shot and compiled up
 > 2.1.101. In looking through /proc, then reading up in
 > linux/Documentation/mtrr.txt, I got the impression that by following the
 > example set forth in "mtrr.txt", one could "...increase performance of
 > image write operations 2.5 times or more." So, I gave it a shot:
 > 
 > 	#echo "base=0x80000000 size=0x20000000 type=write-combining" >
 > /proc/mtrr
 > 	$startx
 > 
 > 	Now for the reason of my posting. Has anyone else done this? I
 > could swear that I _have_ noticed a difference in the rate at which
 > screens are drawn now. Am I just making this up, or does anyone know if
 > this is a true (yet damned easy) way to optimize an X server's performance
 > under 2.1.x? (using a PPro or PII too, of course). Such an easy, speed  
 > optimization seems almost too good to be true, but then again, Linux has
 > amazed me many a' time before.
 > 
 > 	Can anyone confirm that using /proc/mtrr as an easy "quick and
 > dirty" speed optimization for X servers under linux?
 > 
 > 							- Mike
 > ============================================================================
 > Michael Kachline - CS, Georgia Tech
 > kachline at cc.gatech.edu
 > http://brightstar.gt.ed.net/kachline/
 > ============================================================================