Easy way to tell if a seg fault is from ram?

truekaiser · l33t Joined: 05 Mar 2004 Posts: 801

My new build is being stubborn. I have been trying to rebuild @world now that I am on the actual hardware and not the live cd.
I am getting some semi uncommon segmentation faults. And i am trying to determine if it is ram or using -j16.
they go something like

Naib · Posted: Sun Apr 02, 2017 10:12 pm Post subject:

have you checked the voltage?

I ran into an annoying issue when I did a Core2 build years ago .. turned out the mobo was conservatively setting the voltage on the low side while the Giel sticks I was using needed a bit more

found one of my threads from around that time: https://forums.gentoo.org/viewtopic-t-515645-highlight-ram.html

this was also the clock of the ram causing issues BUT I swear voltage was involved as well...
_________________

Jaglover · Posted: Sun Apr 02, 2017 10:36 pm Post subject:

Not being a hardware guru, but I believe theoretically you can have RAM errors if you downclock it, because of less frequent refresh.
_________________
My Gentoo installation notes.
Please learn how to denote units correctly!

Naib · Posted: Sun Apr 02, 2017 10:50 pm Post subject:

Jaglover · Posted: Sun Apr 02, 2017 11:02 pm Post subject:

Those numbers are derived from actual clock frequency, right? Meaning if you lower the clock you also slow down the refresh and may theoretically have a bit flipped every now and then because refresh come too late.
_________________
My Gentoo installation notes.
Please learn how to denote units correctly!

Naib · Posted: Sun Apr 02, 2017 11:08 pm Post subject:

Sort of...
The CAS timings are what needs to be honoured otherwise the DRAM will become ambiguous. If you drop the operating frequency then the CAS settings need to be juggled to ensure the same timings.
_________________

Jaglover · Posted: Sun Apr 02, 2017 11:13 pm Post subject:

Of course rising the voltage should help in case like this, it takes longer for 1 to drop below threshold and become 0 when the starting level is higher.
_________________
My Gentoo installation notes.
Please learn how to denote units correctly!

truekaiser · l33t Joined: 05 Mar 2004 Posts: 801

i forgot to mention those are the default specs of the sticks.
but the asrock taichi board wouldn't post unless i let it auto detect it at 15 15 15 or so at 2133, i had to risk bricking the board by updating the bios to get the xmp profile to work.
Currently testing by compiling kde(want to try it out again) if the compile segfaults again like my first post i am just going to get another set of ram.

I already had to drop 220 to get a replacement board..

Syl20 · l33t Joined: 04 Aug 2005 Posts: 619 Location: France

NeddySeagoon · Posted: Mon Apr 03, 2017 3:28 pm Post subject:

Syl20,

memtest86+ exercises the RAM, CPU and several voltage regulators.
Problems reported by memtest86+ are not always RAM.
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

truekaiser · l33t Joined: 05 Mar 2004 Posts: 801

Well the sticks passed a couple of full passes of memtest86 on the sys-rescue cd. It has the latest version.
I am thinking this may be because I foolishly set the amount of jobs in make.conf to 16. I think I may be getting some bad race conditions because of this but ram is still suspect. Will see if a full system re-emerge will draw out the issue..

Syl20 · l33t Joined: 04 Aug 2005 Posts: 619 Location: France

NeddySeagoon · Posted: Tue Apr 04, 2017 9:23 am Post subject:

Syl20,

memtest86+ hardly exercises the CPU at all, therefore there is no stress on the motherboard Vcore regulator either.
It does do a good job of pushing the RAM subsystem fairly hard.
Maybe Prime95 or cpuburn would find something?

PSU issues normally arise around transient regulation .. for example when the CPU goes from sleep to 95W in 300ps.
I've not seen this for a long time but I used to work around it in older systems by setting the performance CPU governor in the kernel and adding nohalt to the kernel command line, so the CPU ran flat out all the time and the transient regulation demand on the PSU was much reduced.

nohalt is no longer documented in /usr/src/linux/Documentation/admin-guide/kernel-parameters.txt for x86.
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

Roman_Gruber · Posted: Tue Apr 04, 2017 4:51 pm Post subject: