Up to 4.3 Ghz, but at 1.4v

Juddog

Diamond Member
Dec 11, 2006
7,851
6
81
I've been overclocking (or trying to overclock) to get over that 4.2 Ghz barrier this weekend (system has been stable at 4.2 for several weeks) and I'm having a heck of a time getting stable at just 4.3. How safe is it to increase voltage past 1.4v?

Right now I have the CPU voltage set at 1.4v, having passed 9 runs of LinX at 6 GB RAM usage, all 8 cores maxed out. It's watercooled, but temps are hot during LinX testing (hitting 84C right now, but this is after it's been running all night and my room is probably about 82F right now). I think I may finally be stable at 1.4v. If 1.4v is the safest amount, then I guess I'll just try and tweak other aspects. Previous stable voltage for 4.2 was 1.343v.
 

Juddog

Diamond Member
Dec 11, 2006
7,851
6
81
Sorry to clarify, this is an i7 Bloomfield 930, D0 stepping. I have been meaning to post screenshots but haven't signed up yet for any image hosting sites.
 
Dec 30, 2004
12,553
2
76
You need to keep it at 1.343 and 4.2ghz. 1.4v for an extra 100mhz is not worth it. I would be concerned about frying the chip at 1.4v.
 

bryanW1995

Lifer
May 22, 2007
11,144
32
91
it's amazing how quickly vcore goes up when you reach the wall. I was at 1.275 @ 3.88, right now I'm testing at 1.368 @ 4.2. temps up to 83c so far. only needed 1.325 for 4.0, will probably go back there once I get it stable here.
 

Juddog

Diamond Member
Dec 11, 2006
7,851
6
81
It's 1.4v in the BIOS, when loaded up it hits 1.41v or so from the vdroop.

VTT right now is 1.5v, memory at 1.65v.

I guess I could go back to 4.2 Ghz, the voltages were much less, rock-solid stable with 50 passes of LinX at 1.343v CPU. I finally did complete 20 passes of LinX on my current setting last night though, at 4.3 Ghz.

Full settings at the moment, for stable but hot 4.3:
C1E: disabled
Hyper threading: enabled
EIST: disabled
Bclk 205
multiplier 21
RAM x4 mem ratio (1640), Uncore at x8 (3280), forgot QPI offhand but I believe it's around 3600 Mhz.
CPU amplitude at 900mV
CPU voltage BIOS (1.406v), vdroop on, so while running heavy benching ie OCCT it hits about 4.1v
CPU PLL: 1.9
QPI: 1.37
DRAM: 1.65
IOH: 1.25
IOH PCIE: 1.558
ICH: 1.55
Spread spectrum: disabled

During idling at the current setting it's hitting about 43c, during OCCT it hits about 72C max, and after running 20 iterations of LinX it hit 82C max.

When I had it set at 4.2, it was a lot cooler, never got above 71C even after 20 iterations of LinX. I can hit 4.0 on auto voltage which runs a lot cooler, maxes out at 65C or so.

You guys are probably right though, the difference in speed between 4.3 and 4.2 is miniscule for the amount of voltage I have to use, I'll probably back it down to 4.2.
 
Last edited:

Juddog

Diamond Member
Dec 11, 2006
7,851
6
81
:eek: Good god that's a lot QPI/VTT

I'm running an i5-750 @ 4ghz with VTT @ 1.25. I think you could probably lower that?

I started messing around because I was trying to get my memory to 2 Ghz, which I did (well technically 1992 Mhz, which apparently is really difficult to do on this board while overclocking the CPU). Then after finally getting my memory so high, I noticed that my performance in some benchmarks actually decreased, so I lowered the RAM down to 1600 and instead went for tighter timings (right now RAM is stable at 1t-6-8-6-19).

I'll definitely be lowering the voltages, I was just seeing how high I could get it stable at, but it appears 4.3 is the limit right now without better cooling (e.g. running it in the winter, using dry ice on the radiator, using a water chiller etc.) and it's probably wiser to back down to 4.2.
 

SHAQ

Senior member
Aug 5, 2002
738
0
76
Wow...huge increase in temps. Lots of leakage. In gaming 1600 6-6-6 was supposed to be the best. For benching 2000+ at 7-8-7 or less. You can bump it up for a quick bench but I agree it is not worth the stress for everyday use.
 

aigomorla

CPU, Cases&Cooling Mod PC Gaming Mod Elite Member
Super Moderator
Sep 28, 2005
21,087
3,595
126
1.5VTT Good lord!

you will burn out your IMC really quickly at those voltages.

i wouldnt push more then 1.4VTT with 1.5vcore on a bloomfield, and no more then 1.325 vtt and 1.4vcore on a gulftown / westmere-ep.

And no more then 1.275 vtt with a max of 1.45vcore on an 1156.


(disclaimer)
My values only apply if your on a medium tier water setup.
And no all Cool-it water products minus boreas (even then u'll have some problems), Any Zalman water setup, Any Corsair H50, H70, and even if they made a H80 would not be constitute of medium tier.
Unless they made it completely different on the block end, and charged well over 150 dollars for it.
 
Last edited:

Juddog

Diamond Member
Dec 11, 2006
7,851
6
81
My water setup is a 360mm radiator (360mm x 120mm) with two 200 mm fans on top on pull and 3 x 120mm fans below on push. My voltages only got a little crazy when trying to push past 4.2. I was able to reach 4.0 Ghz with most of those settings on auto with the same setup. I guess I was just surprised at how much more effort it took to get past 4.2.
 

Juddog

Diamond Member
Dec 11, 2006
7,851
6
81
Well at least on a positive note, now I know the core speed limit of this chip. I'll start trying to see the limits of the RAM timings next. QPI and uncore now will be tested to the limit instead of messing with the CPU and just leave it at 4.2 where it's nice and cool (running LinX right now, max temp is 68C now that I tweaked my voltages down a little). I switched to auto on pretty much everything except the RAM at 1.65v, left Uncore at auto (shows in CPU-tweaker as 3600 QPI, 4000Mhz Uncore and 2000 RAM at 8-9-8-27,1T). I think I can squeeze better timings from the RAM at 2 Ghz, the 8-9-8 is a bit conservative, since at 1600 I had it at 6-8-6 no sweat - dropping to 6-7-6 started causing memory errors though. :(

I'm less paranoid with a nice 68C degrees CPU versus the 82C I was hitting at 4.3 with the crazy voltages. Lowering the voltage on the VTT definitely dropped down the CPU temps a bit.
 
Last edited:

Juddog

Diamond Member
Dec 11, 2006
7,851
6
81
As an update since I can't get the RAM to be stable below 3T at 2000 Mhz, and I noticed it performs better at 1600 Mhz+ at 1T, I moved the RAM to a 4x ratio and CPU to 20x and have decided instead to try incrementing the blck up higher. From what I have read, this motherboard typically has a stable blck up to 230, so I'm fairly certain I can get back to 4.2 Ghz CPU at 20x ratio at 210 blck. From there I'll just just keep trying to tweak the memory and leave at x4 and x8 for RAM / Uncore.
 
Last edited:

Juddog

Diamond Member
Dec 11, 2006
7,851
6
81
:eek: Good god that's a lot QPI/VTT

I'm running an i5-750 @ 4ghz with VTT @ 1.25. I think you could probably lower that?

QPI = VTT, so VTT was running 1.37v.

I've been doing some research into the matter and am starting to understand better about high BLCK, VTT, and RAM voltage. The higher BLCK basically demands more QPI to function in a stable manner; at 210 BLCK my LinX was failing after 3 runs with QPI at 1.35v, but at 1.37v it's finally totally stable with 20 passes of LinX, and a full hour of OCCT.

It explains the performance of my motherboard now - at default clock, I had absolutely zero issues getting the RAM up to 2000 Mhz totally stable. At 4.2, I had a really really hard time getting the RAM to stick at 2000, even with VTT / QPI @ 1.37. So the higher the BLCK, the more voltage is necessary to keep the RAM at a higher speed. I have read that 1.45v is the max vcore as per the intel whitepapers.

I am really hitting a learning curve when it comes to overclocking the i7's; there are so many datasheets and different forums to read up on to gather the proper information. I also found out that the constant value of 1.65v for absolute top RAM voltage that gets tossed around is typically misused; the actual value is that the RAM can't be greater than .5 higher than the QPI / VTT. So if my QPI is at 1.3, I can safely hit 1.8v max for RAM voltage. Knowing that has allowed me to boost up my RAM to see if I can get some tighter settings, testing it now to see where I can reach.
 
Last edited: