The 22 22 timing at 906mv or even if I go to 950mv in wattman does 2010 mhz max for me which is fine in MEMTESTCL (using memtestcl -g 0 700 99999999
meaning 700 mb over and over till I stop it, 700mb helps me find errors faster than using a higher number)
If I try 1000mv, its less stable. So I wonder if this is some IMC timing limit around 2000 mhz???
Memtestcl is great, it finds errors before even hwinfo reports them! I attached it which includes memtest.cmd--- from cmd.exe prompt call it like this: memtest.cmd 700 for 700 mb for example and it will keep looping it
One cool thing, now when I change clock or volts it does not kill my oclmembench numbers anymore! So perhaps you were right eliovp, the ddr timings spec was wrong and the AMD driver would mess up when you change clock/mv!
NEVER MIND ITS HAPPENING AGAIN, ARGH AMD. Change one thing and oclmembench goes from 228 gb/s to 151 gb/s. Crapola.
Ok found a new fix for that issue. While running memtestcl loop, if I change the mem/vmem it doesn't kill the oclbench scores. Then I can end memtestcl and oclbench/benchmark games keep same performance. Odd, but I guess the opencl thread prevents the amd wattboy (lol) from messing up.
memtestCL-1.00-windows.zip 764k .zip file
Edited by robnitro - 4/26/17 at 8:56pm