Is my i5-2500k the worst overclocker you've heard of?

[Ascaris]

With my Asus P8P67 Pro 3.1, trying to keep the voltage under load to 1.36v, the best overclock I could get 24h stable (I use OCCT large data set) was 4400 mhz. I've tried everything trying to get 4500 stable at 1.36v-- it will not do it.

That is the worst I've read about-- I keep reading about people who think they have a crappy 2500k, and I always end up wishing I was as unlucky as they were.

Last week, my P8P67 Pro decided to quit working while I was using the computer. It's off to Asus for RMA as I type. In the meantime, I bought an Asus z77-a, one of only two 1155 motherboards Fry's had available. Though it has almost all of the OC features in the UEFI that my P8P67 does, it's clearly a budget board more than an overclocking board (as was the other 1155 board).

This motherboard will only get my underachieving 2500k to 4300 at 1.36v. The 4 phase CPU VRM is a prime suspect in this regard-- but it is only 100 mhz lower than my 12-phase P8P67.

PSU is Antec (Seasonic made) HCG-620m. Cooler is CM Hyper 212+.

Just curious as to whether I can at least get the booby prize.

 

[Tradesman1]

I had one a couple years ago that wouldn't do over 4.3, so you're not the only one to miss on the good chips, also ran into a couple that basically capped at 4.4....it happens, it sucks, but it happens 🙁

 

[nikoli707]

my 2500k@4.5 is bios set to 1.340v and runs 1.326v idle and 1.356v load prime/ibt tests. you might be having low voltage idle problems, try using line load calibration and rechecking hwmonitor to see your min/max voltage differences. i will be lowering to 4.4ghz here soon as my ambient temps go up but in winter i will run 4.6ghz at which point im at 1.36v bios and 1.344v idle 1.396v load and i have to use level 4(out of 10) llc to keep it from going from like 1.330v to 1.420v, which i would crash at until i kept raising the vcore until the idle voltage was above a certain level. the llc keeps my min/max voltage closer to the bios set voltage with less of a large difference. i would say your about normal. and were both using a budget hyper 212 so its not going to be high overclocks.

 

[Ascaris]

Nikoli, it's definitely not an issue of idle voltage. I can browse around and do other low-load things forever without problems. However, if I go over 4300 with the Z77-A or 4400 with the P8P67 pro (bclk 100 for both), it crashes under load (stress test) at 1.36v. I've found that IBT maximum and OCCT large data set will trigger a BSOD within minutes at voltages that Prime95 will happily chug along with for some time, and OCCT has the advantage of not trying to roast my CPU in the process. Thus, while I do use a single 10-cycle run of IBT max stress to establish a quick baseline level of stability, OCCT is my main 24h stability tester.

I am using offset voltage, and all I really look at is the load voltage reported by CPU-z or whatever other program (they all report the same for me) during a stress test. If I get a BSOD/freeze/etc., I add a little bit to the voltage offset and retry, until the load voltage hits my desired limit. Then I try to stabilize it by using various other settings (vccio, pll voltage, different VRM settings, etc.,) and see how that goes. I've tried them all, but once the CPU has put its foot down and said it will go to a given speed and no higher, none of the other tricks have worked. I've tried low and high PLL voltages and everything in between, vccio voltages up to 1.15 (memory at 1.5), all of the VRM settings (currently at 350 khz, Extreme phase), c states on, off, auto in varying combos, spread spectrum on and off (currently off)... nothing worked but more vcore (I got it stabilized on the P8P67 at 1.38-1.39 at 4500, but I didn't want quite that much voltage... no one really seems to be sure what the real long-term max is, so I am using 1.36). I like overclocking and have been doing it since the original Pentium days c. 1995, but the reality is that my puter at 4300 performs adequately in everything I do, so I don't want to push it so hard that I wreck a CPU that performs adequately for the sake of having done it.

As long as I don't get idle or low load lockups, I don't worry about the voltage except under load. With vdroop, the voltage is higher at idle than under load (with EISS temporarily off via the "high performance" Windows power profile), but that's fine, as the CPU is not pulling much current at all under low load, and it is current density that degrades CPUs.

I don't look at vdroop as the voltage lowering under load (despite the name that implies this clearly); I see it more as voltage that is at whatever I have set it to under load, and that rises under low loads. Intel knows that lower voltages mean less stability, certainly, and they specified vdroop anyway for the purposes of stability under transient loading, so there may well be some value in keeping it even on an oc'd system. We can get away with less of it (or none) with VRMs and PSUs that are faster in transient response and more robust than the garden-variety ones that Intel knows will be used in a lot of systems, but since vdroop does not actually cause any harm as long as the minimum load voltage is the same either way, why not go with it?

I have not noticed any difference in stability with the various LLC settings with the load voltage set the same. I also have not seen any reduction in the bouncing around of the voltage under load with higher LLC settings. My board has LLC settings of 0%, 25%, 50%, 75%, and 100%, and I have tried them all. Naturally, changing from one of those settings to another will change the load voltage, but after a quick load test, I can see what the new voltage is and correct that back to 1.36. Once that is done, it behaves the same way under load regardless of the settings I used to get there.

At the moment, I have LLC off completely, with the full vdroop enabled. Same load voltage and same max speed as with LLC very high.

 

[bicycle_repair_man]

Wow, reading the comments here makes me grateful that my 2500k can do 4.4Ghz at 1.25V. I thought that was fairly average, guess I got lucky.