[I]Might[/I] be memory; I've had bad RAM cause spontaneous reboots. Check you memory voltage - my 2400 won't work (won't even pass POST) unless the memory voltage is 1.65.

I've *definitely* seen instability with bad memory timings.

[QUOTE=Mark Rose;357018]I've *definitely* seen instability with bad memory timings.[/QUOTE]

So, how best to proceed? Is it reasonable to assume that since increasing CPU voltage by 0.075V has not cured the instability, memory is the likely culprit?

I'm running at 1.65V, try 1.7 or 1.75V? Relax the timings? If so, which of the 30 or so timings should I try changing first?

Intel frowns on memory voltages higher than 1.65 as it can damage the IMC.
Does your RAM have more than 1 XMP profile stored? My 2400 has both a 2400 and 2166 profile available; if that is the case with yours I'd try the lower profile and see if that is more stable.
Let me see if I can RDP to that box and get the timings for the various profiles.

George: Get a bootable copy of memtest86+ and run it overnight at least. It's going to be an absolute cakewalk for you. Just start running it. Take in all the beautiful numbers and letters, walk away and then come back later.

It punches in a whole bunch of patterns into memory and goes back and looks at them later expecting to find the same values. The entire pass is long because at least half of it is spent waiting to see if your memory can hold the bit for something like an hour. If anything is wrong, it will tell you what the error was (i.e. was expecting 7F7F7F7F and got 7F7F7F71 so the error is 000000D) or something, and where it was (i.e. 4598.3 MB) and there's probably something more useful I don't even know how to interpret.

To give you an idea, my 2400 MHz 2x8 10-12-12-31 is perfectly stable for about 14 hours. I got my first error in about eight minutes running with 10-12-11-31 timings. It's [I]that[/I] strong. If you have any memory issues, memtest86+ is going to find them.

As for the cause of the issue, hard to say. Look into loosening timings, adding voltage, look up VCCIO and VCCSA voltages (and maybe PLL, too). Either your chip can't handle the memory or your memory itself can't handle itself.

What is your motherboard? Vendor + model? Is there any kind of over-current protection built into the motherboard?

XMP 2133 pic
My system is stable at 3.9GHz on default voltages (except memory @1.65); CPU HSF won't handle any more than that.

[QUOTE=Prime95;356946]My Haswells (4GHz, DDR3-2400) are running in the mid-70s and mid-80s doing normal LL testing. I doubt temps are the source of my crash-once-every-few-days instability.[/QUOTE]
I'm experiencing similar problems. Perhaps this instability was caused by the nvidia drivers (have a gtx 770). First I'm going to try the new drivers (came out yesterday). If that won't help I'm going to try CPU at stock speed ...
Will post a solution If I will be successful.

Another good test of memory is solving a large matrix with msieve.

[QUOTE=tomaz;357052]I'm experiencing similar problems. Perhaps this instability was caused by the nvidia drivers (have a gtx 770). [/QUOTE]

FWIW, one of my Haswells has an nVidia card, the other is using the integrated graphics processor.

[QUOTE=tomaz;357052]I'm experiencing similar problems. Perhaps this instability was caused by the nvidia drivers (have a gtx 770). First I'm going to try the new drivers (came out yesterday). If that won't help I'm going to try CPU at stock speed ...
Will post a solution If I will be successful.[/QUOTE]

I'm only using integrated graphics (barely even using that - the machine is in text mode and the monitor disconnected most of the time). It's a non-overclocked 4770. I'm seeing hangs rather than reboots - the hung machine is using 50W, rather than the 90W it does flat-out or the 30W idle, and the network light isn't flashing which is what made me suspect network drivers. Though this has happened even while I'm doing linear algebra entirely from local SSD, so maybe the network isn't the problem.

As for a haswell OC guide, I recommend this one:

[url]http://www.overclock.net/t/1401976/the-gigabyte-z87-haswell-overclocking-oc-guide[/url]

I hope it's useful. Sin0822 has written a number of these and looks like he knows a fair bit.

-- Craig