Devil's Canyon - VCCIN/VRIN impact on overclocks.

DubbleClick

Admirable
Following this thread: http://forum.hwbot.org/showthread.php?t=108373

Has anyone had luck with lowering his VCCIN to achieve high(er) overclocks than usual?

I've tried messing with it a little, but it seems like mine isn't willing to work with lower vccin. I'd usually get 4.8ghz with 1.3(1)v vcore on 1.8v vccin, 4.9ghz with 1.37v, 1.82v vccin and 5.0ghz (not exactly stable) with 1.9v vccin.
At 4.8, when lowering vccin to 1.6v'ish, however, I seem to get issues booting windows - once it's booted it works fine, but it can take a few attempts until it doesn't hang. With 4.9 and 5.0ghz absolutely no luck lowering the vccin, not even a slight bit, which is unusual as I even need more than the "golden rule" of 0.4v which applied to haswell. 4.9ghz would still run IBT with 1.65v vccin, but require a whopping 1.43v vcore ().
May this be related to any other voltage settings that somewhat require higher input voltage?
Strange enough, on lower clocks like 4-4.4ghz, I can lower vccin to 1.5v and have it running stable with 1.0v instead of 1.05/1.12 which I'd usually need - possibly even less.
This however doesn't help power consumption, but my chip seems to be very special in that aspect anyway, with 120w power draw during linpack at just 4.0ghz. Testing an i7 4790k on a z97x-sli, by the way. I'll maybe get to try it (and another chip) on a z97x-ud5h-bk and a z97-a soon.

Would be nice if some of you could try lowering vccin and experiment how that affects vcore requirements and post the achieved results.

PS: This doesn't appear to ever work on regular haswell or haswell refresh cpu's. I tried it on an i7 4770k just yesterday but it required 1.85v vccin for just 1.24v vcore (4.4ghz), anything below that it would 0x101/0x124 ocassionally. It would still run with 1.8v vccin with 1.285v vcore (4.6ghz) but crash or hang more often than not. With 1.9v on the other hand no issues occured at all.

Edit: Can go as low as 0.98v vcore with 1.45v vccin (4ghz), anything below that ocassionally crashes in IBT, although other software including prime seems to run stable.
 
Solution
So, after testing the same thing on the UD5H-BK on two I7 4790k's and the I7 4770k I had already tested yesterday. I was able to get 5ghz core, 4.6ghz ring (1.2v vring) benchmark stable, however with a whopping 1.5ghz vcore with 1.98v vccin (something I wasn't able to achieve on my z97x-sli) with HT disabled. It showed no improvement with lower vccins and would in fact not even boot below 1.95v vccin. So it did absolutely not help mine at all.
The I7 4770k saw no improvement with lowered vccin either. 1.9v'ish (only tested in 500mv jumps) vccin were required for a stable 4.5ghz. After all a very strange thing happened and when giving it 1.32v vcore instead of 1.28v, core temps during prime and ibt dropped by 10c. I have no explanation...

DubbleClick

Admirable
So, after testing the same thing on the UD5H-BK on two I7 4790k's and the I7 4770k I had already tested yesterday. I was able to get 5ghz core, 4.6ghz ring (1.2v vring) benchmark stable, however with a whopping 1.5ghz vcore with 1.98v vccin (something I wasn't able to achieve on my z97x-sli) with HT disabled. It showed no improvement with lower vccins and would in fact not even boot below 1.95v vccin. So it did absolutely not help mine at all.
The I7 4770k saw no improvement with lowered vccin either. 1.9v'ish (only tested in 500mv jumps) vccin were required for a stable 4.5ghz. After all a very strange thing happened and when giving it 1.32v vcore instead of 1.28v, core temps during prime and ibt dropped by 10c. I have no explanation for this but it was consistent over 4 times switching forth and back.
Last, the I7 4790k from a friend seemed to benefit the previously required 1.37v required to get 4.6ghz stable with 1.8v actually happened to shrink to required 1.3v (!) with only 1.5v vccin. However, anything below that it would not crash but throttle back to 4.0ghz under load. 4.7 could still not get to run, despite trying to input 1.45v vcore. Lost the lottery, I guess, although his temperatures seemed to be the best amongst the three cpu's rested. The 4770k had the least power draw (110w at 4.4ghz, 1.245v during prime 28.5), but would reach 85c. Friends 4790k had the lowest temps (65c at 4.4ghz, 1.25v) with 120-130w power draw and mine had the highest power draw and temperatures (4.4ghz, 1.1v) with 140w with spikes to 160w and 85c max.

Back to topic, just like in the topic I linked above it's worth to give lowering vccin a try. For the three (two) cpus I got to test, one benefitted from it (0.07v less required) although it didn't enable higher clock speeds. The 4770k seemed to be very vccin hungy to get fully stable and mine starts demanding more than 1.8v when I pass 4.7-4.8ghz.

Sadly I wasn't able to get 5.0ghz with HT running - not even to boot. 4.9ghz (HT on) were stable on 1.42v though on the ud5h, on my z97x-sli I need 1.45v for basic application/game stability.
 
Solution