It definitely does make a difference though, i tested a couple of times at 190x20 (1.25vtt) and 19x200(1.31vtt), (also testing variations of VTT) both failed within 3 tests of linpack (using 3gb). As soon as i tried 181x21 (1.17vtt), it goes on to pass 18 tests before i stopped it. Quite a difference there and it wasn't just a fluke.
But the thing was, at 191x21 (4ghz), it needs a solid 1.376v to at least run several tests of linpack, maybe even 1.392v to be stable. But at 200x20, i can run the tests with 1.36v and needing 1.376v to be pretty damn stable. Leading me to believe the higher bclk along with higher VTT allowed a lower vcore. But this wasn't the case when going for 3.8. This wasn't just once that i tested it either. I also tested it without HT (at both 3.8 and 4.0) and same thing. Just strange.
__
Right, got 3.8 stable. As this morning after leaving WCG running last night, i awoke to a 'uncorrectable hardware error' blue screen. In most cases for me this has been a lack of VTT (which was at 1.17). So i made sure to get it stable before my next crunching session. For some reason the gflops slowly went down after the ~20th test, it was averaging about 49.2 before that.
Vcore under was load was 1.280v/1.296v. Going to see what i can do at 200x19 just for 'scientific' purposes
__
Well, got it stable but it required a touch more on vcore bringing the loaded voltage up to 1.296v. I might try bring vtt down if i can. 200x20 with HT needed 1.33vtt, but due to the lower CPU clock it might need a notch less.
Reply With Quote
Bookmarks