SGI: Hardware

Since there's some interest out there to hack late revision V10 cards into V12's, let me share my knowledge of the EEPROM contents. I personally have no real interest in pursuing this but maybe someone else will take it from here.

Have a look at the L1 output:

So, this V10 board has

(no idea what the T/W is)

The L1 'eeprom' command will dump the contents of one (or all) eeproms in the system. On my Fuel V10 (the blue one), it looks like this:

This part is definitely my V10:

XIO (BIA)
00 09 00 00 25 2f c9 53 4f 4c 45 43 54 52 4f 4e
c7 41 53 54 4f 44 59 42 c6 4d 44 47 38 34 30 cc
30 33 30 5f 31 37 32 35 5f 30 30 31 00 c2 5f 44
01 02 c2 30 30 04 00 00 4c 07 04 00 00 00 02 04
00 00 42 16 c1 00 00 89

My V10 has barcode MGD480 , or hex 4d 47 44 34 38 30 (some bytes are swapped)
It is product name ASTODYB , or 41 53 54 4f 44 59 42

<speculation mode on>both strings are prefixed by hex cN, where N is the length. Scroll back a little and there's c9 + 9 bytes: c9 53 4f 4c 45 43 54 52 4f 4e, or SOLECTRON , the manufacturer!
Next up: cc + 12 bytes: cc 30 33 30 5f 31 37 32 35 5f 30 30 31 => " 030_1725_001 " (the part#)
A '\00', which we shall ignore.
c2 + 2 bytes: c2 5f 44 => " _D " (the rev.)
A '\01' '\02' which we ignore again
c2 + 2 bytes: c2 30 30 => " 00 " (the T/W)

Now, the L1 has an undocumented command 'eeprom fill'. I have never tried it for obvious reasons, but if you want to change the contents of your EEPROM, this may be a start

There's another EEPROM on the card, identified as "XIO (IUA)". Who knows what's to be found there. There may or may not be a CRC on these EEPROMs.

Happy hacking! Please remember, I am *not* responsible for your bricked VPro card, nor a sudden surge of hacked V12's

Thanks j-j!


If some one wants to compare the differences, here's the eeprom dump for the each of the Fuel V12s in my 'graphics-added' O350:
#1


#2

#1 de-hexed in a similar fashion to what jan-jaap did with the V10 data in his post:

jan-jaap wrote: There's another EEPROM on the card, identified as "XIO (IUA)". Who knows what's to be found there.

The "CH" and "IUA" segments appear to be identical between your V10 and and the first V12, but there are differences between the IUA segments of the two V12s.

coming from viewtopic.php?f=3&t=16725878&p=7359377#p7359377 here's the dump of my "exotic" v12:


translators welcome

V10




A broken V12 030_1996_001. If anyone wants to try soldering the eeproms from this board to a 128MB V10, it's available for the cost of shipping - or in exchange for a V12ified V10...

canavan wrote:

Code: Select all

NODE        IP59_4CPU       NTR758         030_1989_003          C    00

If that one is yours: congrats. The 4x1GHz Tezro is about as good as it gets

Looks like this one?

jan-jaap wrote: Looks like this one?

No, it doesn't look like a deskside Onyx2. I thought you'd know that...

canavan wrote: No, it doesn't look like a deskside Onyx2. I thought you'd know that...

copy-paste fail Let me try again: viewtopic.php?f=14&t=12569 ?

Both are purple and highly desirable, though

Hey guys,

I finally had some time to play with the EEPROM's, V10's and a V12.

I have 2 older rev V10's (the 32MB ones), 3 new rev V10's (the 128MB ones), and one V12 (broken - it boots and works, but anything heavier on the display and it freezes the whole system - doesn't even give me POD).

I carefully inspected the boards, and found 2 chips of interest:

Atmel 24C04 I2C EEPROM

Dallas (maxim) 2505 1-wire (with EEPROM)

Interestingly, the cards have pin headers that correspond to those chips - there's a 8-pin header (near the edge connector) wired into eeprom 1:1, 2:2 ... 8:8 etc. (pin number to pin number, NOT in the same order as the chip itself)
and a 2-pin header wired into dallas chip - here the pins are reversed, 1:2 and 2:1 (other 4 pins on the 6-pin dallas are not connected and not used).

I played *a lot* with eeprom reading and flashing (easy), and even after flashing V12's eeprom onto V10's eeprom - the V10 is still a 32MB V10.

What is very interesting, is that the "l1cmd eeprom" DOES NOT show full eeprom contents, it shows the portion directly AFTER whatever checksum and header there is (skipping the first 8-16 bytes or so),
if the checksum is incorrect, everything will still work, but the l1cmd eeprom won't show anything for that device, and l1cmd serial all will report bad checksum.
* I found it the hard way after I realised that I've overwritten a few bytes due to the way my i2c script worked.

Then, I looked at the dallas and was able to read its ID, but I couldn't read the EPROM contents - likely because of pull-up resistor that's somewhere on the board (and I had to use mine too) - that's the specific of 1wire protocol which uses ground and a single wire for power and signal.
Also, the dallas' eprom is *append-only* and it requires 12V programming pulse (which my super-duper device can't produce), I gave up on trying to read and reflash that.

So I took my soldering iron, and swapped the chips, and unfortunately no luck, 128MB V10 with both chips from V12 is still recognised as a V10 with 32MB RAM.

I even removed the small heatsink hoping to find other ID chips, and tried to remove the big one - but gave up, it's stuck on the chip and I don't want to damage a working card.


It is actually possible that SGI did put different chips but used the same RAM because it might have been what they had in stock that time...

canavan wrote: A broken V12 030_1996_001. If anyone wants to try soldering the eeproms from this board to a 128MB V10, it's available for the cost of shipping - or in exchange for a V12ified V10...

Where do you live and how much would the (cheapest) shipping to Japan cost ?

If yours is really broken, maybe I would be able to experiment a bit more (hard to do it on my working cards ensuring they continue working...).

Cheers

kubatyszko wrote: I finally had some time to play with the EEPROM's, V10's and a V12.

I really appreciate your efforts to keep this alive.

If you can work it out, V10 users worldwide will speak the name kubatyszko with hushed reverence.

kubatyszko wrote: I even removed the small heatsink hoping to find other ID chips, and tried to remove the big one - but gave up, it's stuck on the chip and I don't want to damage a working card.

I've had remove the heatsink from a couple of Fuel V12 so the hs could be trimmed to fit into an O350, and the process definitely ain't for the faint of heart. I took some photos of the BUZZ chip last time I did it (pretty sure that was hamei's O350/V12) and posted a cropped version in this thread .

Managed to find the uncropped original on a backup. Dunno if there's anything useful to you in the (slightly) wider view, bet here's a copy:

Thanks, it seems it won't be as easy as it was with ATX PSU conversion though.

For a short time I *suspected* that small 8-pin chip in the upper right corner, I was able to read the markings off it with a lot of light and magnifying glass, but I don't think it's relevant.
The chip on the bottom right is likely a signal converter/buffer, so also a no bueno.

Another important point, is that if SGI was really making them the same with "decision" made after manufacturing, they would likely have had a header, and the only 2 headers are what I already mentioned - likely just for serial numbers.

There is a special many-pin header on the bottom side but that I'm not sure about.

Do we have any volunteers willing to take a heatsink off a V10 (ideally the 030-1826-xxx one) ?