2 Replies Latest reply on Jul 16, 2014 6:08 PM by Cavalary

    Can a CPU core fail subtly?


      Have a Core 2 Duo E8400 which is 6 years and 3 months old. Quite venerable for a CPU, I'll give you that, but I was hoping to get it through to when the next generation of CPUs (and video cards) will hit the market. Also, like the person who posted the topic which is the most recently active here as I write this, computer's on 24/7 and running SETI@home. Actually been running it (with brief experiments with other distributed computing projects) since '99 when it was launched, on all 3 computers I had since then.


      Well, today I noticed that it kept apparently failing tasks, setting them as completed instantly, always on the 2nd task active. A look through my stats on the project site shows a string of invalid results since Jul 8, but definitely not at this rate. Now, of course, the occasional instant or either way very quick finish, quite clearly with an error, happened all the time, but this is definitely something else, not occasional anymore for a week or so and now permanent. So the question is if it may indicate a subtle failure of a CPU core... and how could I test for such a failure in general.

      After some messing with BOINC it seems to work at the moment, but still a few failures earlier, and no idea if the results will be valid anyway. Asked on their forums too, what I did now is based on advice from there, but also saying that if it fails it's probably temperature or failing CPU, and there's definitely still a problem...)


      Additional info, temperatures as reported by Open Hardware Monitor are around 67C / core in full load (so permanently), which is where it tends to stay most of the time, maybe 1-2C less when it's cooler, the difference being the cooler speed (stock cooler), around 2400 now, can be below 2000 during cooler times of year. Also, ran IPDT and that said all fine, but it didn't give me any impression that it tested for something relevant in this respect.

      Also, Gigabyte motherboard, running its energy saving software at maximum level ever since I got the computer. (Incidentally, significant temperature increase if it's turned off. BIOS is set to beep at 60C (external temperature I assume, OHM measuring internal) and if I turn that off with S@h running it beeps in some 20 sec. Otherwise only ever got the occasional temperature warning on days with over 35C, and if I carefully clean the dust when the summer gets bad often not even then.


      So, yeah, sure put it through a lot in all these years, but I want to know what I can do now.

        • 1. Re: Can a CPU core fail subtly?

          It is possible for a single core to fail but even though the Intel® Processor Diagnostic Tool did not report any problem, you may test further by limiting the amount of cores being used; you may be able to change the BIOS or msconfig settings to do so.

          Another option is to assign a CPU thread to a specific Windows* process within the Task Manager’s Processes tab: Right click the desired process, Set Affinity…

          • 2. Re: Can a CPU core fail subtly?

            I gather this means there isn't some software that could just test for such failures, for example by running a few minutes' worth of complex calculations designed to try pretty much everything the processor can do and check whether the results are accurate?


            But it did keep working just fine otherwise since yesterday, very quickly (but not quite instantly) failed 2 more WUs right after uninstalling BOINC, deleting all project data, reinstalling BOINC, reattaching to project and rebooting, but no others since then, so here's hoping I won't jinx it now...


            Seriously dreading a computer upgrade for the past few years, can't even say how long I'd want to wait till making it because I don't see an improvement coming. So far I was thinking I should have somehow gotten something better as soon as I built one, and planning upgrades soon after, the fact that it took several years before actually making a major one having to do with not affording it sooner.

            I mean, there is the one hardware issue I dread to tackle each time, which has everything to do with Intel, namely installing the CPU cooler (embarrassingly always have to ask for help there, found I could do it on others' computers, but if it's mine, I know there's a fragile CPU under there and a bendable motherboard under that and I'm told to apply firm pressure... not happening, my arms turn to pudding, and almost damaged one because of it once... are current sockets any better in that respect?), but otherwise the issue is software, everything more dumbed down, less user-controllable, mobile/tablet/touch/<bleep>-friendly, drivers causing issues, even more so since if I get around to playing anything it's usually an old game. For the past few years I've gone from pouncing on updates and new versions to doing my best to avoid any new version of pretty much anything unless it's absolutely necessary for security, and at times trying to even work around that, and since a new computer would require some new stuff... Yeah, really don't want to.


            (For example, was about to order a laptop last week, then saw it had drivers for Win 8 only... And then that all quad-core Pentium ones I saw were the same. In which case, no thank you.)