Zac67 wrote:
No - the missing address bits caused some memory locations to be blended together: e.g. you write $00000000 to location X and $FFFFFFFF to location Y, and after that you'd read $00800000 from X, because one bit appears shared between the two addresses. The chip has the address bit disconnected and thus can't distinguish between both memory locations.
Aha! I get it! Yes, that makes a lot of sense. I guess it would not be very hard to make a memory tester do "zoned writes", so that each zone has different patterns, and then proceed to read each memory zone and see if it has the original pattern or not. This way only a single instance is needed. Now I wonder if memtest86 does this... (I've not seen something to this effect)
Zac67 wrote:
disassembling the machine (quite a feat on the 3k) I had it narrowed down to a single chip and the exact pins that must have been the problem. I expected damaged trace, bent pins but found - the contacts cleanly suspended in the middle of the hole without any solder. :-P
Hahaha! That's impressive "divide and conquer" problem targeting :-)
Zac67 wrote:
If it's in a vital area, it'll cause a crash quickly, but if it's intermittant, the test software may have difficulties locating it.
When it's in a vital area, is when I call someone "lucky" ;-)
Zac67 wrote:
PS: Just checked on the Viper: the RAM is soldered to the board?? That'll make it very hard to swap for testing and also very hard to find a replacement...
I didn't check the Viper in too much detail, but if it's surface-mount RAM, the trick that I use, which has worked successfully with DIP DRAMs (like those in A500s) is to "piggy-back" a known working DRAM chip on top of the surface-mounted ones and see if that affects the tests. It can be a bit tricky since all the pins have to touch and stay that way until the end of the tests, but it sure beats desoldering ;-)
