Josh Saddler - Latest comments on Failing hardware part 3

In response to: Failing hardware part 3

Branko Badrljica [Visitor] — Fri, 10 Oct 2008 16:55:18 +0000

Oh, one more thing.

If you are byuing new board, ignore boards without SB750 southbridge.

Old ones with SB700/SB600 have various problems and lack crucial function ACC-that might give you better OC or lower power consumption at rated speed and is obligatory for new upcoming K-10 Denebs...

Price delta for such board is too low to not go for it...

In response to: Failing hardware part 3

Duncan [Visitor] — Fri, 10 Oct 2008 15:16:47 +0000

You don't mention whether you tried swapping out or simply down-clocking (if possible w/ your BIOS) your system memory, yet.

I had a 'crashy' system here for awhile due to memory. It was NOT bad memory cells -- memtest checked out fine. Rather, it was that the (generic) memory wasn't quite stable at the rated speed, and would occasionally corrupt a transfer.

As with you, it happened pretty much any time, but more often when there was heavy memory bandwidth usage. My system is ECC enabled and I toggled that on and off, but if anything it was worse with it ON than off, apparently due to the transfer of the additional ECC bits (?). I don't have strong enough 3D to do much there, but I did notice more frequent lockups while doing emerges -- but it wasn't regular enough to really pin it down.

Eventually, I found something on MCEs (machine check exceptions), and enabled the option for that in the kernel. I'd get them, but then had to trace down some way of figuring out what the number meant. I found an app called parsemce (IIRC by Dave Miller?, google...) that told me it was the memory.

Eventually a BIOS upgrade gave me memory de-clocking ability and I found de-clocking the memory a single notch was all it needed. I could even decrease the various individual wait-states/cycle-counts and did so; I just couldn't return to the rated memory clock.

Eventually I upgraded memory and could set the system back to normal memory clock after that. Both the de-clocked old memory and the full-clocked new memory were rock stable, so it was just that, the memory.

If you're lucky, you can de-clock the memory in your BIOS, or have either spare sticks or other machines that you can swap memory with. That'd give you a zero-cost test method. But unless you require registered memory as I did, a single 512 meg or whatever stick should be cheap and enough to test with.

Duncan

In response to: Failing hardware part 3

Branko Badrljica [Visitor] — Fri, 10 Oct 2008 13:18:43 +0000

If you are searching for a cheap 790GX+SB750 motherboard, check out Biostar and their TA790GX series.

They have several very similar submodels, differing AFAIK just by amount of the sideport RAM and one of them being small ITX.

They have had quite favourable reviews and are to my knowledge cheapest in that range.

WHile at it, check your PSU. It might be that bad caps in PSU started it all and that PSU gradually started killing everything else...

In response to: Failing hardware part 3

James Le Cuirot [Visitor] — Fri, 10 Oct 2008 10:22:04 +0000

I was baffled by hard lockups for a year until the only guy I've ever encountered with the same laptop as me managed to figure it out. It was the "processor" kernel module - part of the ACPI stuff. Evidently it's a bug with the machine itself because Vista locks up hard within seconds (rather than hours or days) unless I disable the equivalent Windows driver. It's a long shot but you could try that. It would certainly be worthwhile disabling ACPI completely to see if that makes any difference.

In response to: Failing hardware part 3

Jochen [Visitor] — Fri, 10 Oct 2008 09:51:16 +0000

Hi,

I just want to refer you to my previous comment http://planet.gentoo.org/developers/nightmorph/2008/10/02/failing_hardware#c19978
because the way you describe your problem, the more it looks like you have some power problem (peak voltages). If you don't want to buy an UPS - which I would recommend - you could buy a quality, stabilized (!) power supply which should smoothen the power output.

Cheers
Jochen