i7-3930K: how far can I push it?

[zombie67]

I have a couple i7-3930K boxes. They have hydro coolers. Stock 3.2ghz, and run rock solid at 4.2ghz, with no voltage fiddling. Temps are 54c at 100% CPU load (no GPU to ensure full load). Intel says the max temp is 66c. (http://ark.intel.com/products/63697) So at least temp-wise, I have plenty of head room. Sooo, how much further can I push it and still have it reliable enough for BOINC? Faster crunching is no good if it is down frequently, resulting in less production. I know we have a few team mates with this CPU, and would like to know what they found to successful settings.

[Bryan]

First of all that temperature of 66C is for the "case" not the cores. The cores can go up to 100C and at that point they will start throttling back to protect themselves. Obviously you don't want to get anywhere near that point. From what I was able to find when I was turning on my systems is that 75C and below is safe and anything above that is starting to push.

I found a video on YouTube done by ASUS that showed how to overclock a Sandy Bridge using their MB. Right off the bat they took Vcore to 1.4V and then started tweaking. The goal was to hit 5G which they did. Here again 5G isn't where you want to run your machines 24/7. From what I found you can take the Vcore to 1.4V on air and 1.5V on water. I've never gone above 1.4V and I haven't tried 5G either.

When I built my machines I would turn them on and then run Prime95 for a couple of hours at stock settings to make sure everything was working okay. Then I would set Vcore to 1.4V and run them for 24 hours at 4.8G. None of the machines failed and the core temps never went above 75C. I need to correct that, 1 machine had 1 hot core and would only do 4.7G and stay below 75C.

I am running all machines at 4.2G because reliability is far more important to me since I'm operating remotely. During the last POGs challenge I was in CO and I moved Vcore up to 1.34V and ran the machines at 4.6G for 6 days. None had a problem.

Initially I had a couple of machines that would sometimes give an error at startup and the BIOS would say "failed OC setting" and you had to hit one of the F keys to continue. On the ASUS MB one of the BIOS settings is "Vcore voltage to use at Startup". You should set that to whatever voltage you need to run at the desired OC rate. I've never had a problem once I set that field. If you don't set that I guess it uses "stock" voltage and then sets the OC freq. before upping the OC Vcore voltage.

I just checked and with a 98% load on Ibercivis my cores are running 50-52C.

[pinhodecarlos]

If you guys want to be sure of the stability of the machine you can run LLR that uses gwnum library of Prime95. I've been running LLR since June 2013 in a stable laptop machine and I can give you the numbers and the residues of those ones so that you can compare. If a residue is different then a problem must exist.
Or I can give you a list of prime numbers from different k's from top5000 webpage where k*2^n-1 and when you run them and you don't find as a prime number then you have a stability issue. Also LLR starts giving you round issues on iteration. Please let me know your thoughts.

We could create a thread here with my latest comment where members from here could test and overclock the CPU's based on the result if a prime is a prime number or not. I can make an input file of prime numbers from little size to huge size, like a 48 hour run.

PS( This was how I few years ago I overclocked my CPU's, for a week I would test prime numbers and see if the machine would fail)