A4000 repair troubles only 1M chipmem

[glenn]

Some weeks ago I tried to start my "Main" A4000, in short it didn't start, at all..

I did all the obvious, removed all (7) expansion cards, downgraded from the 060 to a known working original 030-card, removed all memory except chip, ttried with a known working PSU.-. well, everything usual.

After a while I gave up and instead I tried to fix some other broken A4000-motherboard I had laying around so I could replace the one in my A4000.

I have been working a lot with that and I have repaired one 100% that I will use in my A4000, bur now I continued with the original card.

First it did not start at all, I noticed that there was a lot of crap from leaked battery, so i cleaned everything and replaced U177, had to fix several traces with WW-wire, I also had to fix a few signals on the RTC chip, after this the machine boot up, and the RTC even works.

BUT, I only get 1M Chip, the chipmem socket is ok, but the fastmem sockets are broken, but if I hold the SIMM's in place I get 16M fast and that works fine.. not chipmem.

I have..

- Replaced the chimpem SIMM with a know working memory
- Cleaned the socket
- Resoldered the socket
- measured CAS-signals from the SIMM to the destination
- Measured all RAS signals from SIMM to destination
- Mesured all CMA signals from SIMM to destination.
- Replaced all capacitors around the memory, I will replace all of them as soon as I get more, I have ordered everything.

Everything works, except that workbench say I have 1M chip, if I use chuckys Diag-ROM's it also detect 1M, if I do the extened chipmemtest it says that it detect "shadow RAM" after 1M.

Anyone have any ide what to do now ?

..measure the datalines ? ..that really shouldnt give this problem I think ?

..Swap bridgette ? ..I really dont want to if I do not have to..

..Ramsey ? ..but then the fastmem would be affected too ?

..Alice ?

.. U212 ?? (CPLD)

.. U216 ? that is a 74F841, not easy to buy new in sweden.. probably have to order on ebay or something..

Any help would be fine, because I cannot understand why this happens.

[Daedalus]

A simple thing, but it's come up before: What size is the chip RAM SIMM? It needs to be a dual rank SIMM, which usually means only 2MB, 8MB, 32MB etc. SIMMs work. Single rank SIMMs like 4MB and 16MB will all be identified as 1MB by the Amiga.

[glenn]

It is a 2M SIMM, and it show up as 2m in other A4000 motherboards.

It would be nice how the machine find out the amount of memory, does it just try to read it until it end ? some ID in the SIMM ? or ?

[hooverphonique]

Originally Posted by glenn

It is a 2M SIMM, and it show up as 2m in other A4000 motherboards.

It would be nice how the machine find out the amount of memory, does it just try to read it until it end ? some ID in the SIMM ? or ?

Code in the Kickstart will write a pattern and read it back to determine if there's memory or not.

[Franchute13]

Hi Glenn.

If changing the capacitors does not solve the problem, try the following.

Control continuity but make sure that the resistance is minimal. Corrosion of the battery or capacitors can create a resistance on the track/line. This resistance can give you problems.
Upload good quality photos of the battery / memory area.
Regards!

[glenn]

Hooverphonique:

Aha, probably diagROM's do the same, I wonder why it "give up" after exactly 1024K.. or maybe it just look for either 1M or 2M ? ..Most of the data signals is the same as fastmem too, thats why I measured CMA since its different from what fastmem use.

Franchute13:

As I wrote I measured CAS/RAS/CMA lines, all of them are less than 1ohm.

I have cleaned the battery area when I replaced U177 and repaired the tracks to the RTC, so it looks ok now,

I can take a photo later, but I dont think it will say much, everything look pretty ok now after my cleaning and replacement of U177.

[Daedalus]

Originally Posted by glenn

It is a 2M SIMM, and it show up as 2m in other A4000 motherboards.

The 1-2MB region of chip RAM uses the A20 address line. I would imagine there would be other issues as well, but it might be worth checking the continuity of that line between all the major parts (CPU, RAMSEY, Bridgette...)

It would be nice how the machine find out the amount of memory, does it just try to read it until it end ? some ID in the SIMM ? or ?

SIMMs do have a method of size detection that uses a bit pattern presented on some of the pins, but I don't believe the Amiga uses that for the chip RAM detection (hence the problem with single-rank SIMMs).

From what I remember, and in line with the other reply, it writes a known pattern to chip RAM every 4KB, starting low (but not at zero, which is set to 0) and going up the address map until it reads back an incorrect pattern (bus noise/error), or the pattern is found at address zero (address wrap-around). Either of these means (valid) chip RAM ends at the previously tested block. Testing is limited to the first 2MB of address space.

[cpiac64]

http://www.amigapcb.org/ verify connection from socket and bridgette

[glenn]

daedalus: Hmm, A20 ? the SIMM's only have CMA0-9 (and MA0-9 on fastram), and RAS/CAS, I'm not sure hoiw this is "decoded" ?

CMA goes into U216 (74F841) and then continue to Alice.

But logical, if the highest adressbit is broken I should get the problem I have, 1M usable memory and then a shadow of the first 1M over that..

cpiac64: yes, the pcb explorer is my main tool for debugging, it helps A LOT, last time I tried to fix those baords I only had a bad copy of commodores service manual, now I have diag-ROM's, vector schematics and the PCB explorer, it helps a lot.

..I have five A4000 PCB's that ws "too broken" last time I tried to fix them (about 15 years ago), but now I have managed to fix one 100%, and several others now start but with problems. (mainly RTC, memory and sound..)

[Daedalus]

I believe Alice decodes the addresses to the RAS and CAS lines (though it could be Bridgette either, I'm not totally familiar with the 4000), and possibly with the help of a demultiplexer chip of some sort. But the A20 line to Alice or Bridgette could cause that issue if there was some problem at that end of things. If you have a scope it would be worth checking if there's any activity on A20 when it's supposed to be reading in the 1-2MB address range, and also that there's activity on all the RAS and CAS lines. Even if the traces are good, there's always the possibility that a chip has partially failed somewhere.

[Toni Wilen]

Use any monitor program to check if upper 1M mirrors lower 1M or if upper 1M has other weirdness? (for example mon165.lha from aminet, "m <address>")

Exec ChipRAM checks and detects memory aliasing because it is normal when system has only 512k or 1M of chip RAM.

[Chucky]

manual memoryedit in diagrom is a monitor..

[glenn]

It looks like CMA goes into U216 (74F841, transparent latch -listed as obsolete everywhere), after that the signals is named "DRA, and those goes into Alice.

So I guess it's Alice that do the converting. However, since fastmem works fine it shouldnt be any problems with the A1-A22, if it was the fastmem wouldnt work ?

I have measured between the chip-SIMM and U216, but not between U216 and Alice.

So maybe a broken U216 ? or a conenction problem between U216 and alice.. or broken alice..

Or Have I missed something here ?

I have scopes, but not in the same room as I have the computers, but I can fix that.

toni: You mean filling it with data and then see if it is the same data in the second 1M block ?

chucky: I have seen the "manual memory edit" but never tried it.. guess that is the easirst way of testing then.

..I also try to source a SOIC 74F841, there are some on ebay, but none of the "normal" stores have it, like elfa/mouser/digikey/farnell..

[Chucky]

Manual memoryedit is just a editor. you can write what you want to any memorylocation (even diagrom workspacearea.. so be warned!)

so a shadow is simple. if you write data at $400 and have a shadow at $10000 the same data will be shown at $10400

[glenn]

Yesterday I measured every signal between U216 and Alice, no problems there.

..Since I didnt bring a scope I instead started to measuring on another board.. but that didnt go so well either, might start a thread about that too

I also brougth the card to my work to desolder all SIMM sockets, because they are bad.. right now the memory in BANK0,1,2 are glued in place, chip and BANK3 is ok..

[glenn]

I'm not a pro on this, but if I look at the chipmemory first, for example $000FF804, and then look at $001FF804, I see exactly the same data there.

I can also do a manual memorytest from $00100000 to $001FFFFFF and everything is ok.

That really makes it look like that adressbit does not work.

I just wonder why, I have measured everything I think, from the chip-SIMM to U216/U212, from those to Alice, everything is correct.

The only thing I can think of now is U216, that is a 74F841 transparent latch, I have ordered a bunch of those and will replace it.. doesn't sound very likely but I cannot think of anything else.

I also removed all the old SIMM sockets (they where broken and the SIMM's was glued) and soldered in a brand new one for chipmem.

..And to add to the chaos.. suddenly when I was running the memory test something in the audio area started to make smoke! ..not much but a little, and now the audio volume is really low :/ ..have to look into that later, probably an OP ? dunno.. but I repaired audio in A4000 several times before.

[glenn]

When I get home I will carry the scope to the computerroom and see if there is any activity on A20.. but im not sure what it would tell me, it seem to be obvious that that the problem is there somwhere..

I wonder when I get my 74F841's from poland

[glenn]

I brought the scope, and looking at the A20 is ee no activity.. however, I see no activity at all on A1..20 when scanning memory ?

[Daedalus]

Hmmmm... Where are you checking the lines? Different systems have their own address buses so in many cases you'll only get activity on the relevant bus. The CPU address lines should be active during RAM checking though.

[glenn]

I checking at the fastslot, it was the easiest place I thought.