Originally Posted by dirtyred
AVX is generating quite alot more heat so I would start with an offset of 2-4. If you're not using softwares daily with AVX instructions then there's no use of it anyways. For gaming you will see absolutely no benefit for it (yet).
Try manual fixed voltage. RealBench is doing fluxuating load to simulate real life usage and causes frequency and voltage fluxuation which could cause instability at certain frequencies where the voltage is not enough. Voltages are not linear to frequencies. If it's stable with manual, then go back to adaptive but add a really small offset, like 0.02-0.05 V and lower the turbo voltage by the same amount giving you the desired total Vcore. This last step resulted in stabilizing my 4.8 GHz with AVX offset. At 4.8 GHz it was stable but once AVX was detected it dropped to 4.5 GHz but didn't get enough voltage and crashed.
On my board and probably on others LLC level 5 gives the closes Vcore to the set value under load. Setting IA AC / DC load line to 0.01 should help with more stable voltages.
If unstable but you can't increase voltage more because of heat, try lowering RAM speed and voltage (try 1.2 V between 2133-2800 MHz with default timings or even a bit looser). That should decrease temperatures by 3-10 degrees. Instability can be caused by too much heat as well because of electromigration.
You can lower cache frequency to 47-46 and see if it's more stable. Cache speed doesn't result in much gain anyways. See guide in first post or you can benchmark it for yourself.
Personally I just keep cache at 42. It doesn't hurt. IF you're trying to dial in your CPU overclock, keep everything low for now.
With that said, if one drops AVX by 2 - 4 how does that affect Prime 95 benching? Does Prime 95 then look more in line with other benchmarks? I'm curious because if you drop AVX sufficiently, could it be possible that other benchmarks now generate more heat than Prime 95?
Originally Posted by becks
Agree on that, but as I said I think
the v are actually lower due to Wattage used and lower temp, at same freq with same test on the same PC
but yet again it might be just margin of error..
Still testing, the trouble is that I work so much ...I only have like 2-3 h / day to put my hands on it...that's like 2 RB runs and 3 bsod's and that's it
Will report later how its doing but still need any advice that might help me go 5.2...5.3 without going over the top with volts...maybe i'm missing something in bios...
I think you should always measure load voltage versus frequency rather than measuring what you set as different motherboards behave differently in terms of LLC and what not. Ultimately the most accurate would be using a DMM or oscilloscope to measure the vcore each time your CPU is tested for stability. I personally don't do that so just measuring load vcore vs frequency vs temperature is good enough for me.