I think it's good in all situations to have LLC 100%. My chip is very unhappy when there is vdroop. It may be a chip-to-chip thing, but it seems all the heavy overclockers use it.
Seems the opposite for me.More gflops at 2ghz 9-9-9-25-110-1T mem with 3.6ghz uncore than 1600mhz 7-8-7-24-88-1T mem with 3.2ghz uncore. I tried this in diagnostic mode last night, also. Still ~60gflops with all mem.
I played with vQPI up to 1.425v last night. 1.4120v allowed me to get my uncore up to x19(3.8ghz), but I couldn't hit that magical 4ghz. The step from 3.2 to 3.6 only needed .0125v increase. The step from 3.6 to 3.8 needed .062v! Almost 5 times times more!!! I think I need 1.48v+ for 4ghz uncore with my memory.
1.425v and pcie freq of 104 allowed me to fully stabilize 210bclk tho! I don't think I want to run 20x210 for 4.2 instead of 21x200 tho... I don't think the difference will be that much, and the added volts bring my temps up from 84c max linx to 88c max.
Things I still want to do some more:
Retest with higher vQPI while dropping ram to 2T. This should help with uncore stability, I think?
Try to get linx stable with 4.5ghz HT Off just to get into the 4.5ghz club. There was a 990x shot of 25x180 doing 91Gflops with 6/6 instead of 6/12! My goal is to beat that with 6/6.
Also thinking about getting an external rad box. Maybe one of those 9x120mm deals. Was also considering like an xspc radstand. Plumb those things in with some QDC's so I can switch it in and out of the loop easily. For 24/7, the triple 120mm rad setup I have is fine, but it just lacks the capacity required for those higher clocks. I see everyone doing 4.5 with like 70-80c, and I'm doing 4.2 with 80+...
Bookmarks