Slashdot Mirror
Software To Diagnose Faulty PC Hardware?
Etylowy writes "Over the years I have repaired my own PC and those belonging to family and friends many, many times. While in most cases it turned out to be restoring a system after malware/the user/Windows made a mess, or simple cases of 'follow the smell of smoke and molten plastic,' there were some nasty ones where the computer mostly works. By 'mostly,' I mean: you can boot it up, it might even work for a while, but will crash way too often to blame it all on Microsoft — what do you do then? Once you strip it of any extra hardware (which, with today's motherboards that have pretty much everything integrated, might not be an option) you are left with the CPU, motherboard, graphics card, RAM and HDD. You can test the HDD, you can run memtest86+ to check the RAM, but how do you go about testing the CPU, motherboard and graphics card trio to find which is to blame? Replacing them one by one isn't really an option. Do you know of any software that would help the way memtest helps with RAM?"
The phoronix test suite is a good profiler, at least it would narrow the search. But, as you observed, once you are down to the RAM and integrated devices what options do you really have expect to toss the mobo?
self-checking programs like Prime95 can be useful to test the computer more generally (if you've verified with memtest a failure here basically means cpu/chipset at fault).
Other things I've tried before have been (if the motherboard allows) things like significantly underclocking sections of the motherboard/processor, if an specific underclock fixes the problem you just significatnly narrowed down the list of possible failures.
there are similar programs to memtest that will check a GPUs output conforms to what it should, but if you just have random-crashy-badness that can be a pain to diagnose. Sometimes things like just running without graphics drivers for a while can help spot those problems, if the computer no longer crashes you can look a bit further away from the graphics card as most of it's capabilities won't be used.
Well... typically you find the fault by using an application which stresses one of those components far more than any other and then seeing if the failure condition you're observing occurs more often. This is just basic troubleshooting, it's not even specific to computers.
#fuckbeta #iamslashdot #dicemustdie
That's not an Online UPS, so it won't protect against all grid issues. And Online UPS are expensive and noisy.
Hiren's BootCD contains a bunch of different utilities for doing just this. Plus it's bootable, so if you can't get into the OS you can still use the CD. It can do just about anything you'd need to in order to diagnose and repair a machine. You just gotta find it (usually the pirate bay or other torrent sites are a good place to look.)
Even when they do, it's usually a sign the rest of the board is on it's way out too. A device on the board not functioning can mean a number of things (MB controllers acting up, visible/non-visible corrosion in the board, blown capacitors, etc), so you can be up for a lot of weird behaviour from the board that you can't pin down.
To be honest, relying purely on a test suite to tell you what's broken will lead to disaster. Only through experience do you get the pointers toward what is actually faulty. Add to this that true diagnosis only comes from swapping out parts, and, well, test suites don't look at all like a viable option.
When I am repairing hardware about the only suite I use is memtest86+ and a decent live linux distro. You can usually pick devices that have failed with lspci, however this is not always correct. It all goes back to having test hardware & the knowledge of what certain behaviours in systems are caused by certain faults. After 15 years of working in IT with both hardware & software faults, there's only so much you can do with limited or no test hardware. Most of the time when you're diagnosing hardware faults on the phone it's an educated guess at best, the only time you truly get a decent diagnosis is when you have the machine with you and can swap parts out. Hell, we don't even use the Dell diagnostics at work due to their inability to give decent results on anything other than RAM.
Check supply voltages first.
There's a really fancy test program to do this... it's called a digital multimeter, and it's a piece of hardware with two probes.
You touch one probe to ground, and then use the other to check all the leads going into MB for supply voltage.
For desktops that is.
For servers, the power supplies are generally smart modular units, and you check their voltage outputs in the BIOS screens, or using remote management via BMC: IPMI, iLO, Drac, or ALOM
While that is good "Bad or Maybe" test, most PSU problems are transient over- or under-voltage conditions, which a DMM is not going to reveal.
And there are testers that will measure all (or most) of the voltages produced at once - you jut plug the atx cable into the device, and many of them have a pass-through, so you can test the PSU under load. I'd look for one that could flag a transient problem, if it exists.
Mind you, since writing the above I have looked around for one, and have failed! They all are pretty simple devices that do not detect transients, I could find no pass-through devices, and they all test under very anemic loads. All told, I am not impressed by any of them.
Prediction for end of Universe #42: Fencepost error in Quantum_bogosort.cpp
Yes... but unless you are doing this professionally, or going way out of the way to build a full blown test rig and load bank yourself, the gear required to fully test a PSU anywhere near max load is not worth it to the average person, a spot check with a DMM on the bench or in the PC (if the PC is working) is a good tradeoff, and if there is any question, try replacing the PSU.
Versus buying a $100,000 Sunmoon or Chroma tester. Or bench Oscilloscope + DC Load generator + Variable AC output gear (for varying input voltages to the PSU under testing). To be honest, all this sort of gear is pretty cool, and would let you even get an idea about how clean the output signal is from the PSU, it's just overkill to do that much testing as end-user for PCs.
On the other hand... no geeky technology enthusiast should really be without a Fluke DMM or similar piece of gear in their bag of tricks with at least ability to measure emf, current, resistance, and (maybe) LCR, over the years i've found it indispensable.. measuring electrical characteristics is useful for many things, not just for PSU testing.
I won't knock the little ATX test products, but they're really no better than a DMM and a big resistor.