Hi!
For my two 5870 @ 930 / 1300, I'm getting:
GPU_USE_SYNC_OBJECTS=1 DISPLAY=:0 ./cgminer/cgminer --scrypt -o
http://MY_IP:9327 -u ltc5870.2 -p X --shaders 1600 -I 17
~300kH from each
NiceIf I run with: GPU_MAX_ALLOC_PERCENT=100, I see:
[2012-07-30 16:21:05] Error -5: Enqueueing kernel onto command queue. (clEnqueueNDRangeKernel)
[2012-07-30 16:21:05] GPU 1 failure, disabling!
GPU0 seems to work with GPU_MAX_ALLOC_PERCENT...
Any better setup for 5870s?! On top on Ubuntu 12.04 64 bits, Catalyst 12.6 and SDK v2.6.
Tks!
Thiago
Clocks are too high I had similar until I started playing around after noticing a 5850 of mine that will only go to a core of 765 was getting the same Kh/s as my 5870s high clocked so now I use 770,1050 and get 356Kh/s on Ubuntu 11.04 64 bits, Catalyst 12.6 and SDK v2.7. Oh and forget the --shaders it does not give you the correct .bin file used it is always lower than if using --thread-concurrency 8000,8000 in your case which will give you a 8000.bin file used as opposed to on my install somewhere in the 7000s.bin it was using with the --shaders. The startup file I use below the 5850s seem to like 8192 as opposed to the 7200 it should be for 5x shaders that seems to be the sweet spot for all my other cards this on the new 2.6.1 code I have the pools set in the .cgminer.conf. The -I 18 gives me the same amount of stales as the -I 17 does but get a few more kh/s out of the cards 19 or 20 give a little higher speeds but results in double or triple the stales respectively so in the end leave you with a lesser overall effective speed.
cat ltc.sh
export GPU_USE_SYNC_OBJECTS=1
export GPU_MAX_ALLOC_PERCENT=100
export DISPLAY=:0
~/cgminer-ltc --scrypt --worksize 256,256,256 --thread-concurrency 8000,8000,8192 --vectors 4,4,4 --gpu-threads 2 -I 18,18,18 -g 1 --auto-fan --auto-gpu --temp-target 81
My method for getting the most from the cards.
1. Start with --thread-concurrency at 4x and 5x shaders with an 8192 thrown in to see what gets you the fastest speed.
2. Move onto the core clock raising it up/down until you get the fastest kh/s the core is usually always lower than what you think it should be.
3. Now on to the memory clock raising it up/down until you find the sweet spot that gives you your total highest speed you will get, this again is usually lower than you think it should be on an over clock.
4. Play around some more after you have what you think is the highest speeds as sometimes that extra/lesser core/memory 5-10mhz will find a new sweet spot for you resulting in an extra 5-10Kh/s. Always go with an ending number of 0 or 5 as the others seem to screw things up.
. Since my computer is running on Linux and I'm a noob in linux, please don't ask me to send you debug files or anything like that 
. I didn't have "bitforce:" line for every device btw, if that helps.
