Anyone using linux to benchmark the stability of their OC'd processors?

[ApHtEr]

Currently I am using HPL (Which what IntelBurnTest) for stability testing and I am using Stream to watch my memory bandwidth.

Running CentOS 6.3, ... was just curious to see if anyone else was running linux around here and what their OC'd were like.

I've come to notice that Windows seems to be a little more forgiving on high clock speeds than linux especially when it comes to really heavy benchmarks like HPL.

--Tony

 

[Plimogz]

Compiling the kernel is one extremely harsh test of stability -- so, if that counts towards your definition of "using linux" to stress test an OC, then, yeah.

As for the particular app (HPL) that you mention, I have no experience. My closest experience is using Prime95 on linux to stress test before prolonged distributed computing duty and from my experience linux was indeed "tougher" on the OC -- but the difference was rather minimal, say 50 MHz -- interestingly the Linux systems also turn in faster times than the their windows counterparts, so it sort of makes sense to me.

My guess is that Linux has less overhead (or something, I really don't know) so it can allocate more cycles to the actual work and thus increases the rate at which errors might crop up.

I base this solely on the fact that my stress test would error-out sooner on Linux correlated with the fact that a Linux system returns more results Vs. a Windows box.

 

[Kalessian]

I have had a very stable PC pass all the usual tests but fail compiles of big software (ie openoffice). Kernel wasn't a problem, just the 24h compiles would fail until I reset to stock.

 

[JQuilty]

I generally consider my OC's stable if they can transcode a Blu-Ray movie in Linux using Handbrake. I use Fedora for this.

 

[PreferLinux]

I've used Linpack, so probably the same as your HPL. I found that I would hit a point where the system would lock up – it seemed fairly stable (a couple of runs of Linpack) with a lesser OC.

I can't really compare my results to Windows as I have AVX on Linux and not Windows, but I know they seem fairly high compared to other results I've seen. I do know that I get higher results if I test without a GUI / X running.

Compiling the kernel is one extremely harsh test of stability -- so, if that counts towards your definition of "using linux" to stress test an OC, then, yeah.

As for the particular app (HPL) that you mention, I have no experience. My closest experience is using Prime95 on linux to stress test before prolonged distributed computing duty and from my experience linux was indeed "tougher" on the OC -- but the difference was rather minimal, say 50 MHz -- interestingly the Linux systems also turn in faster times than the their windows counterparts, so it sort of makes sense to me.

My guess is that Linux has less overhead (or something, I really don't know) so it can allocate more cycles to the actual work and thus increases the rate at which errors might crop up.

I base this solely on the fact that my stress test would error-out sooner on Linux correlated with the fact that a Linux system returns more results Vs. a Windows box.

I agree Linux has less overhead and can therefore spend more time doing the actual work. That obviously increases performance slightly, but because of the extra performance temperatures (stress) will go up. That means you OC is "less stable" than it would be, and therefore more likely to fail. A side effect is as you mention that errors can show up sooner as its getting through the work quicker.