So for months i was running 1000core and 400mem, Getting a happy 320mh/s per 6870, I have two, And they are in a gaming rig.I Have figured out the error!
After a clusterfuck of driver changing.
MSI 2.1 gave 0's when "readingout" my gpu stats,
So i upgraded to 2.2, And i got proper readings.
With the mimimum memclock being 575<-normal
So, As usual i enabled the unofficial overclocking... I dragged the slider to 400, Hit OK, It said "OK" but the mem clock Always goes to 575, Regardless of what i select.
So i went "great, MSI 2.2, And Cat 12.3= no 400mhz memclock"
I decided to use CGminer to control the clocks, But to no avail...
I got the standard "GPU something reports OK for clock change, Check clocks to make sure"
So i check the clocks, And 1000core, 575mem....
Im wondering why i cannot bring my 6870's memory clock down anymore... before the driver cluster fuck i was using msi 2.1, cat 12.3
Turns out one of the cores on the 6990 was defective, And was crashing and Halting, But not inturrupting much of anything other than readouts and anything that attempted to use the "crashed" core would cause a bsod, (I had not installed anyform of auto restart)
Why was CGminer failing on my desktop?
Well that one was high OC+heat related and is totally solved with a can of air now. For whatever reason CGminer's interface likes to lockup when instead it should flag a HW error.
Why the hell was GUIminer working on the "crashing" 6990 and CGminer was not?
I cant really answer that since i dont know... But perhaps gui miners poclbm has someform of crash reduction, Or like a "less harsh" mining style, Or a auto restart, Etc.