i7-4770k Random Instability

spaceman1701

Distinguished
Nov 1, 2011
91
0
18,630
I've got a modest overclock of 4Ghz on my 4770k and I've been trying to under volt it some because my cooling solution can be very loud and I'll be using this settings as my day to day setup. I was able to lower the voltage to 1.060 volts and the processor seemed stable. It passed a ~3 hour prime95 torture test (using small FTTs). I didn't have time to run prime95 any longer, and after 3 hours I figured it was probably stable. A while later I decided to do some gaming, and everything worked fine for about 3 hours. After that I got a BSOD.


I've raised the voltage slightly and once again, it seems stable, but it seems strange that at 1.060v it worked for so long before crashing. Should I just have run stress tests for longer, or is it possible something else is effecting the stability of the system? I've never had an Intel CPU before, and I've heard Intel's power saving features can be problematic when not running at stock voltages. Could those be the problem and, if so, How can I disable those in my BIOS (I looked but couldn't find anything that was obviously power saving features)?

Thanks for any help.
 
Solution
You probably weren't stable at 1.06.

If you want to be 100% certain you are stable, you need to blend test with Prime 95 at least 18 hours (it usually takes that long to run through all of the tests-there are ~60 of them) in addition to other basic tests-a huge pain and stress on the CPU.

I find it more sensible to do P95 custom 90% memory (or IBT would work too)~1hr, then small ffts ~1hr. If it passes, run a gaming benchmark and/or play a resource intensive game. Finally, I'd run P95 blend for a few hours.

If your Haswell is like the SB CPUs, power saving is often called "CPU C1E", "C3", "C6", and "Package C states." Better to turn them off when using offset voltage, but they don't make much difference otherwise.

bebop460

Distinguished
Jun 25, 2011
169
0
18,710
You probably weren't stable at 1.06.

If you want to be 100% certain you are stable, you need to blend test with Prime 95 at least 18 hours (it usually takes that long to run through all of the tests-there are ~60 of them) in addition to other basic tests-a huge pain and stress on the CPU.

I find it more sensible to do P95 custom 90% memory (or IBT would work too)~1hr, then small ffts ~1hr. If it passes, run a gaming benchmark and/or play a resource intensive game. Finally, I'd run P95 blend for a few hours.

If your Haswell is like the SB CPUs, power saving is often called "CPU C1E", "C3", "C6", and "Package C states." Better to turn them off when using offset voltage, but they don't make much difference otherwise.
 
Solution

spaceman1701

Distinguished
Nov 1, 2011
91
0
18,630
Thanks for all your help!

Thats very low for the frequency that your trying for id keep bumping it up till you get no unstability
Okay. I'll do that.

You probably weren't stable at 1.06.

If you want to be 100% certain you are stable, you need to blend test with Prime 95 at least 18 hours (it usually takes that long to run through all of the tests-there are ~60 of them) in addition to other basic tests-a huge pain and stress on the CPU.

I find it more sensible to do P95 custom 90% memory (or IBT would work too)~1hr, then small ffts ~1hr. If it passes, run a gaming benchmark and/or play a resource intensive game. Finally, I'd run P95 blend for a few hours.

If your Haswell is like the SB CPUs, power saving is often called "CPU C1E", "C3", "C6", and "Package C states." Better to turn them off when using offset voltage, but they don't make much difference otherwise.

I'll keep that in mind, and I'll try what you suggested for stability testing.