Diagnosing A4000 instability (RAM?)

[zero]

I am having some stability issues with an A4000:

Cyberstorm MK3 060/50, 32MB on board
8MB on the motherboard
GVP SCSI card, 4MB on board
Buddha IDE card
Picasso IV

I'm not 100% sure about the RAM sizes, but basically the motherboard is full, the Cyberstorm has a couple of SIMMs and the GVP has one SIMM.

Sometimes programs crash or the system reboots with a guru. One consistent way to make it happen is to run any of the memory test progs I found on Aminet (memtest, memtest2, memdiag, memcheckBH, bustest etc.) They run for a bit and then the system freezes and resets.

Is there some way to diagnose this issue? Other than just removing SIMMs one by one, which could take ages... Assuming it even is RAM, if it was you would think that the memory test would find it rather than crashing.

[Daedalus]

Well, you could remove all the SIMMs and see how it runs on just chip RAM, then add half and try it, then half again (or remove half) for the minimum number of changes. The problem could easily be with one of the SIMM sockets either as they're notoriously fragile. Also, have a look at SysTest by Keir Fraser. It's a bootable ADF image that doesn't include the OS in its memory testing routines and will highlight any bits that produce errors, letting you narrow down the location of the fault.

[NorthWay]

Can you reproduce it exactly the same way every time?
I'm asking as you hopefully have made sure it isn't any buggy program that trips you up.
If you suspect this to be the case I would add debug tools, of which the MuForce and Guardian Angel would do (or others, depending on what 060 library you use).

But a physical memory check shouldn't be too much trouble either, just take the binary search approach:
- Take out half of the physical memory in the machine. If it still crashes then repeat.
- Add back half the memory you have that was taken out with the last crash. If it doesn't crash then repeat.
- If the last set of memory you put back in isn't the smallest change you can make then start from the top again, but only go about halving the chips you just put in.

That should be good for finding one fault. If you have multiple then you will need to do it all again.

[zero]

Thanks Deadalus, I'll try SysTest before starting on the SIMMs.

NorthWay: It's repeatable with multiple test programs, so I don't think it's a buggy app. I run them after booting from cold with no startup-sequence so that should rule out any drivers etc.

[zero]

Okay, I eventually found that if I removed an 8MB SIMM from the SCSI board (actually a DKB Rapidfire) the memory test progs run fine. System seems more stable now, sometimes it would crash if you did "assign env: ram:" but not any more.

Still can't get Comadose to work for some reason. I could have sworn it worked on my old CS 060 MKII.