SGI: Hardware

Sorry, now added to the post above!



I hope so!

One (or more) of those pins should go to GND, which should be common with the ground lead shown in the pinout (unless Dallas did something rather strange).

My bet from looking at the datasheet is that the +95mV is the battery-in lead and the +58mV is the Dallas chip's output to keep the SRAM valid (with a bit of voltage drop from the chip's switching feature).

It must be what I've labeled B and D, then. Although it's not a direct short to ground: resistance is around 160k ohm between those points and pin 14 (GND).

okay, here's the deal:



your mission is to figure out which pins on the dallas chip are battery in, VCC in, and RAM VCC out.

BATTERY -, CARRIER GND, and RAM GND should all be the same.

as suggested earlier, the higher voltage you're seeing is probably the raw battery output. find the same voltage on the dallas chip pin(s) - hopefully it only goes to one place. the lower voltage is probably the output of the switch/rectifier in the dallas chip, and it should also be present on pin 28 of the socket to feed the RAM.

if the higher voltage does indeed go to one place on the dallas chip, find a trace to cut to that pin, or nip the pin as close the circuit board as possible. after doing that, make sure the lower voltage you found is no longer present anywhere.

if that happens, you've removed the old battery from the circuit and can solder the + terminal of your new battery (it is 3.3V) directly to the dallas pin you cut.

now prepare for the joys of non-Y2K compliant clock chips - you're on your own there!!!

Amazingly, my PowerSeries (4D/380, 4D/440) know about post Y2K years, but the Crimson doesn't.

As long as you run a Y2K patched OS (IRIX 5.3) on them there are no issues with the old 4Ds. On IRIX 4.x and older, the day of the week doesn't match the date post Y2K

yes, the the question is whether the clock chip itself rolls over correctly at the end of the year.

you'll have trouble if:

1) the year rolls over while the machine is off, or
2) the year rolls over while the machine is on, but the software doesn't write the new date to the chip on the assumption that it works correctly.

it shouldn't be a big deal to fix the date once a year. and the next question is whether the counters in the dallas chip actually support (have enough bits to store) years > 2000.

Here's what I've found are the relationships between the exposed contacts and the pins on the little chip itself (the one inside the carrier surround):


The highest voltage relative to carrier GND is at the small chip pin 4.

I'm hopeful that we've isolated the battery + to the contact labeled A and small chip pin 15. However, there isn't enough physical space to get any type of clipper in to cut and bend the pin. I'd have to try to separate the two plastic bits of the carrier, and it looks like that might destroy the whole carrier. Hmm.

Those aren't the only (small chip) pins with a voltage on them.

ok, looks like if the A,B,C,D labels on the chip pins correspond (check with an ohmmeter) to the A,B,C,D labels on the circuit board pads in your picture, then you can hopefully just cut the trace stub leading "up" from the A pad. the assumption is that battery + is wired to that pad underneath the carrier board, and cutting that trace will disconnect the battery from the whole circuit. use a SHARP exacto-type knife, being careful not to cut any other traces (or yourself).

try it. you can always bridge the trace with a tiny bit of wire or a solder blob to put it back as it was if this doesn't work as expected.

after making that cut there should be no battery voltage anywhere you previously found voltage, except at the isolated A pad. if that's the case, hook your new battery + to chip pin A or the stub of the cut trace (NOT the pad).

remember we don't want to hook the new outboard battery to the old battery. the new battery would attempt to charge the old battery, at best greatly reducing the life of the new battery, and at worst making the old battery leak or catch fire(!)

The other thing you can do (asuming a plated-through hole) is take a small drill and scrape away the top of the plating where the hole goes through. Then you have the whole pad to attach a new wire to (a zot of nail polish can keep the solder from bridging the gap to the hole again).

The good news is that it is possible to remove the upper plastic surround from the main carrier assembly without doing irreparable damage:


The bad news is that there's no single 'smoking gun' pin to cut and re-route....



Yes, that's how I verified that they were connected (an ohmmeter).

Okay, well, I've tried it. However, cutting pin 15 still left power to other pins. The highest voltage all along, even before the snip, was at pin 4 and did not correspond to A, B, C, or D. So I cut that one too. Well, several pins still have a potential difference between them and carrier GND, most notably 14. So, I *could* cut that too...but we're obviously getting into a situation where it's not a simple battery-to-pin connection I'd have to replicate -- it's a whole circuit of some sort, unknown at this point.



For reference, the magnitude of the potential differences was greatest at pin 4, then 15, then 14. This means that the battery was most likely to be applied with the least amount of intervening circuitry to pin 4, etc., in that order.

We may have to go back to the diode plan, since I can't see how to proceed here with three different voltages -- obviously the true battery + connection is obfuscated by/behind circuit board.

At this point I'd go back to attacking the epoxy with either a solvent or heat.

ARGH!

you should have tried cutting the trace leaving pad A first!!!

So far I didn't have the opportunity to run tests with different MC2 boards. I did find one which I'll post a high resolution picture of as well as two additional old pictures from the archive:

030-0117-002 Rev A

Date: 11-08-90



This board is equipped with 2MB SIMMs. I've compared the other two to this one and there don't seem to be many differences.

030-0117-009 Rev A

Date: 09-05-90



I'm not sure about this board, the only 009 MC2 I have in my list is equipped with 32x 2MB SIMMs. Compared to the one above I can't see any differences. An interesting point may be that it is older.

030-0117-003 Rev B

Date: unknown



According to the list it is equipped with 8x 8MB modules which would match the picture. At least this board is different from the other two. Most obvious is that they use a different style of sockets now (I think these have metal clips). Also, the packaging of the ECC controllers is different (still Rev 1 though). Less obvious is the higher revision on some parts of the chipset (middle row: DECODERH, DECODERL, DECODEWH, DECODEWL upgraded to 070036x00 3 ).

I don't think anything I've done so far is irreversible. I can reconnect the pins and cut the trace if you think it's worth it.

That just seem counter-intuitive! Is it possible that when SGI did upgrades at client sites, part of the deal included turning in older boards, which were then factory-refurbished or something? Could that explain anything?

Also, in some other SGI parts, the zzz of xxx-yyyy-zzz in part numbers don't necessarily represent increments on the same design -- they might differentiate completely different boards.

well, we're trying to isolate the built-in battery's + terminal from everything else. it's quite likely the A pad (with the highest voltage reading) is it. cutting the trace coming up from the A pad would tells us. if after the cut the A pad is the only place you have voltage, then you're done and can hook your new outboard battery to the side of the cut away from the A pad.

True, but:
(1) the A pad it only the highest voltage of the pads, not of the small dallas chip pins
(2) the A pad is cut and there are still voltages differentials between the pins and the carrier ground, so cutting both A and the pin with the highest voltage have not produced the desired results. That is to say, I measured between cuts, and just cutting A did not remove all deltas.

Which, I think, leaves us with either the dremel method, the less-ideal SRAM (only, not timer chip) on battery method, or getting a new dallas assembly.

I finally had the chance to run a test with a 030-0117-001 MC2 board and as you describe, the 8MB SIMMs are not recognized. With 8 SIMMs installed the PROM reports 16MB and no error on startup. IRIX boots fine, for very low values of fine (5.3 isn't fun with only 16MB).

The other equipment in the system I used is rather new (IO3B and IP15). Given that it normally runs with a newer REV MC2 (8MB SIMMs installed) I presume it is safe to say that the limitation is related to the MC2 board itself.

Compared to the pictures I posted earlier the 003-0117-001 MC2 looks quite different. Most obvious is the lack of labels on most parts and a lot additional blue patch wires.

Great -- I'm glad we've narrowed it down now. Thanks for taking the time to check out the MC2!

I've made good progress with the battery replacement. I found an online chip distributor which sent me a sample DS1216B package. I put it into the IO2 board and attached the original SRAM chip to it. The PROM now remembers settings between power cycles! I haven't checked into the system clock yet, but that's next on the list. Here are some pics of the battery replacement process.















She's ALIVE!