Overclocking Intel i5-760: IMC voltage limits

[dizcza]

I overclocked Intel i5-760 2.8 GHz to the next stable status (stability was checked by running a blend test in prime95):
150 Hz x 22.0 = 3300 GHz
Core voltage: 1.25 V
IMC (VTT) voltage: 1.4 V
Another voltages (like PLL) I set to auto.
When I became setting IMC voltage > 1.3 V, its value turns in red color (which is bad, I think). I read some forum posts about max safe IMC voltage is about to 1.35...1.4 V. Then I understood, that bigger IMC voltage leads to bigger temperature and reduce CPU lifespan. The first is OK (t <= 65 C during running a torture test in prime95). But the second... I don't know how to figure it out.
The questions are:
1) Will my CPU be (likely) working after 1 year under that OC conditions? Or it turns out, that after a few weeks I'll have to buy a new one? I have been using this CPU for 2.5 years under the stock condition (no OC).
2) If it will, why don't put IMC voltage to 1.7 V, which is max, and overclock CPU to 4.x GHz, If under this condition it passes the torture test?
3) In some posts I saw my problem can be solved by decreasing IMC voltage while increasing CPU Core and PLLvoltages... so sweet things it might be! I did try a lot to adjust these voltages but my PC didn't load Windows. Is it real to adjust CPU Core/PLL/PCH voltages to decrease IMC voltage?

Added: My motherboard is Asus P7P55D-E... I read that these motherboard series have mmm not very good chipset for overclocking under safe voltages... maybe I'm one of losers who bought it)

 

[Burpo]

You didn't say what board you're using, but it's better to keep numbers out of the RED. Increasing core, pll & pch volts is normal when overclocking, you just have to find that balance between power & speed. As long as volts aren't too high, the CPU will last just fine (years)..

 

[Deders]

1.4v does seem a little high, what voltage is it running at stock?

Can you try lowering the BLCK and see what VTT voltages you need to have to boot windows at lower speeds, say 140MHz, 130MHz etc?

My i5-750 isn't anything too special in terms of voltage to clockspeed ratio but I only need 1.212v on the VTT to make it 100% stable at 3800MHz (180MHz BCLK) and it's been running for about 4 years like that.

I'm also wondering if the memory itself is affecting it. The only time my overclock started giving me errors in Prime 95 was not long after I had to RMA my sticks back to OCZ because they didn't run at advertised speeds. The replacements needed a slightly higher VTT to run stable.

Have you got load line calibration enabled?

 

[schmuckley]

bah..should do 3.8 easy
Not a super-bad board or anything..but yeah.
Here ya go Miahallen style.
http://www.techreaction.net/2010/09/07/3-step-overclocking-guide-bloomfield-and-gulftown/
Dude was an 1156 wizard.
I has some experience clocking Lynnfields
Eh..920s do better..period.

 

[Deders]

Come to think of it, the VTT is only supposed to even out the difference in voltage between the cpu and the memory so there isn't too big a jump between them. Too much voltage would be as bad as too little voltage. (or possibly worse). Remember with these chips if your Memory's voltage is more than 0.5v away from your VTT then it can damage the memory controller.

have you tried it at around 1.2v or 1.25v?

 

[dizcza]

Thanks for replying. But the thing is... I am not a luser but rather have fingers are all thumbs! I've overclocked my CPU today to the 175 x 21 = 3675 Hz using the next voltages:
CPU Core: 1.367x V
IMC: 1.30 V
Memory: 1.625 V
I don't know how clumsy I was to not pay attention to Core voltage... I did my wrong conclusion (about only IMC voltage is matter) because I didn't try to crank up enough Core voltage to pass an overclocking. Sorry for that.
But now I face the problems of :
1. Overheating. I use the biggest radiator for my PC box and MX-4 thermal compound (both are good as I think).
The Intel i5-760 specifications say that the maximum allowed temperature is 72.7 C. Prime95 blend test heats up my CPU core to >=75 C using the voltages above... and I think it is too much for me, isn't? Not to mention that today it's too cold that I wear a pullower in my room... the summer kills my CPU!
2. I set CPU clock ratio in BIOS to 22.0 (22.0 is max; 21.0 is stock) to have 175x22 = 3.85 GHz... but CPU-Z and RealTemp shows me that I use only 21.0 ratio! What the f... When I decrease BCLK to say 150 Hz, setting clock multiplier to 22.0 is absolutely OK! Why is that and how to handle it?
3.

Come to think of it, the VTT is only supposed to even out the difference in voltage between the cpu and the memory so there isn't too big a jump between them. Too much voltage would be as bad as too little voltage. (or possibly worse). Remember with these chips if your Memory's voltage is more than 0.5v away from your VTT then it can damage the memory controller.

Well, this is interesting because in my case IMC V. < Core V. < Memory's V. If VTT (IMC) voltage is supposed to be in the middle between Core voltage and Memory's voltage, I'm in a bad shape... or not?

bah..should do 3.8 easy

... not for me

 

[Deders]

That looks much better. Still a little high but so long as it works it looks fairly safe.

MX4 is great, my temps are 60c full load with Prime64, but my cooler has direct contact heatpipes, but only 6 of them. One of the things that really helped was intake airflow, moving the disk trays from the front of my case so the fans had more direct airflow. What case do you have?

How did you apply the MX4? I find thinly spreading it with a card works best for me. Also it needs to heat up for an hour or 2 and cool off for a while afterwards before it becomes fully effective.

These CPU's are actually ok up to 99c but I wouldn't be happy if it was going above 80c, that being said it would rarely reach that during gaming even if it did with Prime95.

Download Argus Monitor so you can accurately see exactly what your CPU is doing. In the performance tab with the bar graphs, if the top of the bar goes orange then it is using Turbo to reach your highest clockspeed, this will be the extra multiplier.

I don't think the IMC necessarily has to be in the middle, it's certainly not on mine, it just needs to be raised a bit as you raise the Core voltage.

 

[dizcza]

Answering to your previous question about Line Calibration, it's on.

What case do you have?

I use Chieftec SH-01 case. But i put it between a wall, a wall and my table so there is small intake airflow comes into the front of the case... i know, it's the worst scenario to cool the PC but it's irreplaceable.

How did you apply the MX4? I find thinly spreading it with a card works best for me.

I applied MX4 to the radiator (Zalman)... Its connective layer is a bit rough so I must put a thick layer of thermal pasta to cover all CPU area. I think it works fine.

Download Argus Monitor so you can accurately see exactly what your CPU is doing. In the performance tab with the bar graphs, if the top of the bar goes orange then it is using Turbo to reach your highest clockspeed, this will be the extra multiplier.

I do see only one orange line... Thus, the extra multiplier will be added as soon as all my CPU cores loading riches 100 % and it won't be enough to handle some task to CPU so he turns on an extra multiplier?

 

[Deders]

I'd try a thinner layer of paste, a thick layer will just act as insulation and trap the heat. The idea is to fill microscopic holes.

Basically if turbo boost is enabled then it you might as well consider whatever frequency it reaches on all 4 cores when stressed your overclock figure in MHz.

For instance my overclock is 19x180.5 but it turbo boosts to 21x180.5 which is pretty much 3800MHz.

 

[dizcza]

Okay, Deders, thank you very much for feedback! The rest of the CPU OC I'll do on my own.
The next step is to overclock the GPU... but It'll be another topic

 

[Flapdrol1337]

I could run my i5 750 with the 21 multi (stock 20), but to do that I had to turn on the turbo and disable all the c states. If you set windows to high performance it'll stay at max multi.

If you keep c states on the cpu will try to turbo to 24 multi or something when on a single core and that won't be stable.

I had to increase one of the "other" voltages, not by much and I don't know what it was called, anyway, with that I could run 200 fsb, of course I had to lower the memory multiplier.

 

[BonzaiDuck]

. . . The Intel i5-760 specifications say that the maximum allowed temperature is 72.7 C. Prime95 blend test heats up my CPU core to >=75 C using the voltages above... and I think it is too much for me, isn't? Not to mention that today it's too cold that I wear a pullower in my room... the summer kills my CPU! . . .

This is a case where some respondents dropped a minor ball.

Maybe there's still confusion about it.

The 72.7C Spec is the "TCASE" spec. The TCASE spec is a general thermal guideline that is supposed to drive computer case and cooling solutions, and would seem to insure that the silicon isn't submitted to temperatures that might cause more damage over a longer time.

The TCASE spec is not based on a "real" sensor; it is instead a temperature taken at the IHS processor cap's dead center. Measuring TCASE directly is not an exercise that enthusiasts would pursue, unless they wanted to cut a channel into their heatsink base, or some other possibility that could capture that temperature.

Tom's Hardware some years ago -- just around the time of the Wolfdale C2D release -- published an article about this. Measurement leads to a conclusion that TCASE can be somewhere between 10C and 15C less than the average of core sensors depending on the TDP of the processor and possibly the total number of cores. The difference between TCASE and an average of Ti -- core sensor temperatures for each core i -- the 10 to 15C -- is determined by the thermal properties, size and thickness of the nickel-plated copper IHS.

This is my understanding of it. If by some . . oddball chance that I'm wrong about this -- correct me.

So core average temperatures might be as much as 83 to 88C for that spec to be exceeded. What it is exactly -- I can't tell you. But the safest thing I can imagine without totally abjuring use of overclock settings would be to stress-test up to a temperature of ~83C as measured by the core sensors. Even safer -- apply that guideline to the maximum single-core temperature -- ignoring the fact that there is error or miscalibration of the core sensors -- sometimes put at between 5 and 6C either way.

If it seems too complicated, think about it long enough and it should make sense.

 

[dizcza]

Measurement leads to a conclusion that TCASE can be somewhere between 10C and 15C less than the average of core sensors depending on the TDP of the processor and possibly the total number of cores. The difference between TCASE and an average of Ti -- core sensor temperatures for each core i -- the 10 to 15C -- is determined by the thermal properties, size and thickness of the nickel-plated copper IHS.

Wow, amazing! Seems like a foundamental fact for all CPUs. Can you give me a link where it has been discussed or when it's been tagged as a true conclusion?

 

[BonzaiDuck]

Wow, amazing! Seems like a foundamental fact for all CPUs. Can you give me a link where it has been discussed or when it's been tagged as a true conclusion?

That I would like to do, but I'll have to search for it with no guarantees that it's still there. You could, however, find the detailed specification explanations for that particular processor -- or for that matter -- any intel processor. There is an entire section (again -- "as I recall") devoted to "thermal" issues, and TCASE should be discussed there. That's where I also recall reading the information. In the Intel material, they simply explain TCASE -- what temperature it represents -- where that temperature would be measured.

But I'm sure there was a Tom's Hardware article about it.

UPDATE: I found this at "Tech Reaction" dated 2009:

http://www.techreaction.net/2009/10/14/guide-to-understanding-intel-temperatures/

This explanation controverts or changes the one I had given from hazy recollection of the previous readings, except that it was my understanding from readings at that time that there is . . no . . . TCASE . . sensor. It also insinuates that the temperature is measured at the center of the heat-spreader, but UNDER it.

The article may even link the TH article I had read before, but I get that "Error 404" when I try to open that link.

What I can only assert with confidence is that the article I read before put the difference between the TCASE and Tj sensors in that range between 10C and 15C*[see * below].

Here's another forum exchange at TechPowerUp which deals with it, and it seems to confirm some part of the previous link. http://www.techpowerup.com/forums/threads/questions-tcase-temps-on-the-i7-920.128453/ But it also substantiates my point: the processor won't throttle at "TCASE Maximum" until the Tj temperatures reach the other [Tj] spec for the processor -- implying the difference between the two.

Now -- if you look at the new Haswell-E processors, you can see that the TCASE spec is about 5 or 6C lower than the spec for -- say -- my Sandy Bridge K. It's around 66C versus 73C. This might reflect how the die size and number of cores affect the difference.

I'm known around here for a conservative approach to my over-clocking. I won't run the voltage beyond either a spec upper safe limit -- or a consensus about an upper safe limit when Intel no longer publishes one (they DID -- for the Nehalem 32nm processors, which as I understand was the last time.) When my Tj core sensors are showing close to an average above 80C, I don't push it further.

But I'm quite sure that any TCASE that can be measured is always less than what the Tj sensors would show under identical conditions. And I'm quite sure I had read -- in a TH article -- that it would be a difference of between 10C and 15C for the processors of that time*.

What has changed since then is the lithography (from 45nm to 32nm to 22nm) and the die-size along with it. Does this mean the difference is smaller for post-Nehalem/post-Sandy-Bridge cores? Or larger? That -- I can't say. But I think the processor of interest to you here is the i5-760 -- a 45nm processor following Nehalem for socket-1156. The lithography matches the cores referenced in the article that I can no longer find. So the 10 to 15C approximation would apply with greater accuracy than it would for my 2600K chip.

 

[Deders]

These Lynnfields will throttle at 99c, but like I mentioned before 80c is the max temp I'd be happy a chip reaching.

 

[BonzaiDuck]

These Lynnfields will throttle at 99c, but like I mentioned before 80c is the max temp I'd be happy a chip reaching.

That was the "half-gen" CPU line between Nehalem and Sandy, and it had the 45nm lithography of the earliest Nehalem cores -- as I understand it. So as I've tried to argue, it would fit into the context of what I'd read in the now-hard-to-rediscover article.

But look around at the current-gen landscape now, and the threads popping up here. Whether attributed to the AVX2 features of stress-tests, the TIM employed in non-E cores like IB and later, simply the gap between IHS and die, the smaller lithography or die-size -- it is harder to keep these chips cooler with conventional means. Yet operational voltages are lower.

For instance, on that particular Lynnfield, there was a "safe-range" voltage spec upper limit of 1.40V. The 32nm Nehalems were spec'd at 1.37+V. After that -- no published spec for the 22nm processors. Another forum member -- IDontCare -- posted in threads showing that a desired clock-speed was more sustainable at lower temperatures with slightly less voltages, and that conversely -- increased clock-speeds and temperatures resulted in disproportionate power consumption with greater electrical leakage.

Then, look again at the TCASE spec for the Haswell-E cores. TCASE for the non-E i7-4790K is ~74C or 1C higher than for my i7-2600K. But TCASE for the octo- and hexa-core chips is ~66C. And at certain overclock settings, the Haswells seem to just reach a wall of instability -- less like the steeper side of an exponential curve and more like a "discontinuity" or vertical line.

Maybe it's just my imagination. Or too many threads with people reporting high thermal levels, even with voltages that seem otherwise sane. One could suspect that the temperature target for overclocking these new CPUs might just as well be the TCASE spec mis-applied to the core sensor values, and not something above it.