QPI inverse scaling, and final ghz acheived
edit note > make sure you case is well ventilated ie no heat build up overtime inside case and you have the best cooler you can afford mine was a megahalems for this guide.
hmm, ive got to be honest and say that to prove absolutely tangible evidence of inverse core temps with higher QPI would be pretty arduous, and im thinking perhaps i should now take a break from OC testing and install some stuff, especially as the volts various i7’s are happy with seem to vary allot, so im guessing proving it will have to wait until ive got the time, to run some very scientific tests. but certainly the message from all my experimentation seems to be certainly for my setup that when QPI/VTT is AUTO set, i seem to be able to get to the higher speeds with much less trouble, as I’m currently stable at 4.2ghz im probably going to stop there, i’m writing this running prime quite happily at 4.2ghz full load for the last 2hrs, and she passed IBT (Intel Burn Test) on maximum for 3 passes without being above 82c on any core, I’d say IBT is the only stress test that can properly test stability to its extreme over the short term (5-10 mins) and that prime95 serves only for long term stability testing(hours, days etc), under prime95 blend at 4.2ghz its not going over 78c max on any core, PCH 32c and CPU 56c wierdly with this all my settings are auto ! apart from QPI PLL set to 1.28v (stock 1.1v) and CPU PLL set to 1.88v(stock 1.8v), one thing that worries me is now and again hardware monitor records a minimum cpu fan speed of 0rpm for a second or something, but this is another issue im sure this must be a software glitch but im going to check over everything, full fan spin downs seems to be 15 x longer than this 0rpm glitch appears for as I only see it ont he minimum column of hardware monitor and never catch it in the value column itself, so i think im ok eek. apart from the QPI pll and CPU pll being manual
this is what the bios is choosing to use volts wise for the other AUTO settings :
so settings in resume for me are :
bclock 200mhz and cpu multi 20x (turbo enhance enabled which pushes to 21x ie 4.2ghz stable)
this setting above might be whats making 4.2ghz a doddle on my machine ? as 4.2ghz starting with a 21x multi seemd much harder to acheive stable style.
also EIST is currently disabled not sure this is helping will turn it on at some point, so CPU-z reports my processor as running at 4200mhz constant as turbo seems to be permanently on in this scenario ?
Hyper threading is off (i figure iwas pushing my luck with HT at 4.2ghz on air ? another time ill test though i dont think HT is worth the temp increase it yields compared to higher OC.
QPI multi is 36 x
so QPI seems very happy at 7.2ghz
on the ex58-ud3r bios Performance enhance is set to extreme ?
auto sets CPU vcore to 1.4v in bios but in software ranges from 1.360v to 1.380v (stock vcore1.2875)
load line calibration is on, CPU-Z observed under full prime95 load it seems to run at 1.36v constant.
auto QPI/vtt is ending up at 1.515 (stock QPI/VTT 1.175)
auto set IOH core to 1.3v (stock 1.1v)
auto sets ICH I/O to 1.5v (stock 1.5v)
auto sets ICH core to 1.1v (stock 1.1v)
auto sets DRAM volts to 1.6v with my ram set at 1200mhz and 6.0 spd
auto sets PCIE to 1.5v
manual QPI PLL 1.28v (stock 1.1v)
Manual CPU PLL 1.88v (stock1.8v)
nb if pushing RAM to 2000mhz dram v goes upto to 1.7v but at 4.2ghz im running memory at a 6x SPD at 1200mhz this helps allot ?
all these auto volts settings were viewed/discovered with gigabyte easytune6 booted in windows ? so im guessing since this a gigabyte motherboard
theyre very accurate ? I know auto OC settings for everything other than QPI PLL and CPU PLL goes against all the grain of OCing law ? but my setup when set like this seems more stable than any other way, I have set it up, and it would be interesting to see if some people who had problems with 4ghz and above could try the same approach and see whether it worked for them. obviously you do this at your own risk, and most in the OC community would say that QPI should never be as high as 1.515v, ie 1.35v max for QPI/VTT ?, and they might be right as im willing to risk it on the basis im prepared to risk the effects as 77c max and the fact that auto set QPI/VTT when running 1600mhz and 2000mhz ram at stock CPU 2.66ghz speeds always seems to be 1.5v for me anyway : so can it be so wrong ? but see what your board auto sets QPI/VTT to when you run 4ghz and above or with memory at 1600mhz and above ? and reply to this thread post with your discoveries as i would love to know ? youll need easytune6 to view these auto volts settings from windows so im guessing only people with gigabyte boards will be able to do this ?
my MB is a EX58-UD3R v1.6 bios FG (listed wrongly as 1.0 earlier in thread)
Im going to upgrade the bios soon im hoping it doesnt throw all my settings out otherwise ill have to retrograde the bios, as these setting took too long to find for me perosnally to lose them for a BIOS update, but hopefully the BIOS update wont change the functionality of any of these Auto v settings.
I also decided to underclock the RAM ? as this seems to make it more stable at 4.2ghz I run my stock 1600mhz ram at only 1200Mhz ie the minimum 6 x SPD ram multiplier with a 21x cpu multi at bclock 200mhz, as this seemed to work for me perhaps its a trade off, between running the ram high or running the CPU 4.0ghz and above as at the moment i cant work out how to get both, as the major cause of instability for me seems related to syncing high mhz memory with the bclock set high enough to get 4ghz and up and when I do push higher with ram speed the auto set QPI/VTT seems to go higher still upto 1.55v ? but this doesn’t seem more stable at this point.
so a recommendation from this memory experience would be if 4ghz and higher is proving difficult for you, clock your ram lower it should help, though i wouldnt go below 1066 mhz as this would be a bit nuts.
also if 4.2ghz is proving difficult try a 20 x CPU multi with bclock 200 with turbo enhance on (and possibly eist off ? maybe speedstepping frequencys at 4.2ghz and down is more difficult for the CPU to handle the v fluctuations successfully ? but this is just what mine happens to be and works i have not tested whether speedstep is causing issues)
now that ive come to the end of OC’ing my i7 for a little bit 4.2ghz is more than i could of expected, i must say its been an interesting experience, thanks for all the help in getting to this point, and I hope this post proves usefull to someone in getting 4ghz and up.
improvements to test
I will report back once ive turned speedstep back on to determine whether that creates any instability issues, if not ill probably have that on as running constant at 4.2ghz seems a bit energy ineffecient though idle cores range between 40-46 after 3 hours on.
my next step might be to set these auto settings manually and see how stable it is with them set manually, and perhaps to then push lowering vcore to try and prove this inverse theory ? but at the moment I dont have time to go this far.
Wierdly with these auto settings all set manually i thought it seemed less stable which I cant really explain ? the reason why i decided to manually set QPI PLL and CPU PLL is they seemed to be the two settings that auto didnt raise in any way ? as i went higher with the bclock, and this is only a theorisation but since it seemed to handle raising all the other values well by itself, that perhaps these apparently static values on auto were getting out of sync with the rest, so i reasoned to raise them up, and this seems to help with stabilisation for me.
I do wish there was some totally clear scientific explanation of how all these volts settings inter-relate with each other and what ratios need to be preserved for stability, but i guess only the Intel i7
engineer who overclocked his own PC could tell us ? I wonder what speed he’s running at and what he understands about the i7 that the wider community doesnt ?
cheers to all at Techreaction