SGI Indigo R4K Failure – Let out the magic smoke! Burning smell

SGI Indigo Failure

So I *finally* managed to find the keyboard cable for my SGI Indigo on my last trip to Norfolk, VA. This completed what I needed: the 13w3 to VGA cable, the special keyboard and mouse, the system and the keyboard interconnect cable. I humped the thing onto my desk after clearing a bit of space, hooked it up to my monitor, then switched my first SGI on. It powered on! But then came this nasty smell. That’s not good! It did show the clock TOD error, no chime sound (just clicks.)

I originally bought the Indigo from Reputable Systems (Greg Douglas) in the 90s. I was working tech support at the time, living with my parents, and it cost a good chunk of cash — especially by todays standards. Originally it was R3K with a lower level graphics subsystem that I believe could do some hardware 3d acceleration but only 256 colors. I’ll admit when I first started messing with it I was a bit disappointed. It was still sort of slow, but had good audio capabilities and worked well as an X-term. It wasn’t long before MP3 was a new thing and the system couldn’t decode a 128kbps MP3 real time though, it took about 6 minutes for a 4 minute song — so I would have to write it out then write it to minidisc. Creative commons licensed music, of course.

Somewhere along the lines in my SGI fandom I think I ended up with a 9u VME frame that had an ELAN graphics boardset on it. I also ended up with a 2nd SGI Indigo from NASA auctions I used to frequent, one with the faster MIPS R4000 series CPU (R4K.) I remember posting that I was giving away a bunch of hardware including parts from the NASA Indigo, and left it and other stuff behind my office in Virginia Beach for someone else to pick up. But a homeless scrapper got to it first, taking everything metal and leaving plastic shards of laser printers and SGI Indigo (and other equipment) behind. Ugh.

But I held onto my Indigo, not having looked at the CPU cards in a long time. The first thing I did was pull out the CPU card and thought, “man this is bulkier than I remember! What a crazy heatsink for this slow R3000 processor.” But it turns out the R3K board isn’t a R3K board, it’s a R4K board. So then I’m thinking, “Okay so I guess I kept the R4K CPU and the lower graphics board.” But no, the graphics boardset looks stout. So I guess I kept both higher end parts and gave away the old R3K / Entry stuff. Or maybe the Entry graphics was in one of the bigger deskside systems on the frame the Elan came on. I can’t remember.

I remember that the Indigo had two power supply options. One for R3K/Elan graphics and one for R4K. I thought you could use a R4K with entry graphics on the weaker power supply as well. This was a huge issue in my mind since I figured my Indigo had the R3K PSU. I pulled the PSU from Indigo (My Indigo was always named Eggbert from a Super Mario Bros 2 character of similar name that was purple, so from here on it’s Eggbert.) So Eggbert’s PSU turns out to be the R4K model! I opened it up, the capacitors look okay but that really doesn’t say much. No burnt smell inside PSU.

So back to sniffing the circuit boards more. I sniffed and sniffed. Two times over I missed the roasted capacitor on the bottom of the CPU card. I was thinking it had come from the graphics boardset, which sits in the case right next to the bottom of the CPU boardset.

The fried capacitor is C531, and this is near the sound section of the CPU board. No telling what it actually goes to but the lack of the chime sound might have something to do with it. Image slideshow above has a picture of the fried capacitor. Audio sections of computers often use the negative voltages for opamps and DACs where the rest of the computer might not use these voltages, so I might have to check to make sure the power supply isn’t doing something bad on the negative voltage outputs (if it has them.)

In the process of diagnosis I have made a list of all similar style capacitors on the board. I don’t know what is wrong with the machine, but if this style/make of capacitor is degrading from age then the solution should be to replace them. They’re SMD, but pretty large SMD so replacement shouldn’t be TOO bad.

So here is a list of the capacitors I have made so far. I’ve only researched 2 of the 4 with regards to what standard the size is. I have what I believe are DigiKey part numbers for the value that fried and the most common other value (prob used for noise filtering on the power rail.)

I measured the size with a caliper and best guessed on what it equates to in industry.


SGI Indigo SMD tantalum caps of the style of the one that burned up:

Top of PCB:
C14, C15, C23, C48 = 47uF,16v
C30 = 33uF, 20v

Bottom of PCB:
C555, C544, C525, C513A, C513B, C578A, C578B, C645A, C645B = 47uF, 16V
C697 = 33uF, 16v
C551 = 33uF, 20V
C507? = 10uF, 10V

C551 is the crispy one

47uF/16v = 7.3mm x 4.42mm : Size 2917 (7343 Metric) / Digikey – 478-1739-1-ND
33uF/20v = 7.44mm x 4.33mm : Size 2917 (7343 Metric) / Digikey – 399-3788-1-ND
10uF/10v = 5.5mm x 3.3mm :
33uF/16v = 7mm x 4.45mm :


I will also have to do a battery holder mod since TOD battery is dead.

More updates once replacement caps come in and I replace them.

Coming soon: RGB LED Video Screen Expansion

RGB LED Screen Expansion

Thanks to Supersoaker and Eric! Supersoaker laid out and cut 4 wood panels that I will use to add two square meters of screen to the RGB LED video screen I built before. Once designed this cuts down on the amount of time it takes to cut out the backing that goes behind the LED modules. Screw holes, data cable holes and power cable holes were cut. Each screen section is 1/2 a meter high x 1 meter wide. This was done at the NovaLabs makerspace, a place I should probably get more involved in (after MAGFest!)

I still need to pick up a miter saw and cut a few other boards, then I should be able to assemble the screen, connect it to my existing panels and program the control boards.

More to come on this one!

GeForce FX5200 overhaul, Andamiro Mark 6.1 / VI – Pump It Up Arcade Hardware – Fan, Swollen Capacitors

Andamiro Mark VI / GeForce FX5200 overhaul

A few years ago I got a decent deal on a dead Pump It Up NX2 cabinet (Thanks Matt!) I was originally after a DDR machine, settled on a Pump It Up machine, then got a DDR machine while waiting for the PIU deal to go through. Both machines were dead, and needed to be repaired. I suppose they’re always in a state of repair, but that’s okay.

A few weeks ago I let my friend Greg borrow the computer from my Pump It Up machine to do some software development/QA on newer versions of the software that run on these machines (he does contract work for the company that makes it.) While the computer was out of the cabinet, I figured I would replace the cooling fan on the video card. It had seized up before I got the machine. It still runs but it’s probably only a matter of time before disaster hits in the form of DRAM failure or GPU failure.

So while the computer is out I ordered a replacement fan from China for the NVidia card:
Triangle Screw Distance: approx. 25mm between each two
Mounting Hole Dia: approx. 2mm
Dimensions(mm): approx. 36 mm(Diameter)

(See slideshow pics above for pics of the video card.)

These haven’t arrived yet so I don’t know how they will mount up. The fan on my heatsink is removable via 3 small screws access through the blades.

While the card is laying on my desk, I look at it and notice bulging capacitors. Common issue on a lot of electronics, not surprised. To be fair it might be from the excess heat from the GPU not having a running fan, but while we’re rebuilding this thing might as well do it up right!

Caliper measurements on the caps says they’re 8mm x 12mm, 1000uF 6.3V

A good replacement seems to be Digikey Part: P5509-ND Panasonic 105c with 3.5mm lead spacing, 8mm diameter vs 12.5mm height .

Going to add this to my DigiKey order, will update this post with results.

I had to replace the power supply in this computer already. For those that don’t know, EVERYTHING IN THESE PUMP IT UP CABINETS RUN AT 220V. Yes, it’s fed 110-120V but they run a 1:2 ISO transformer!!! Fair warning, if the PSU isn’t auto switch on the computer set it to 220/240v!

And no, I’m not very good at Pump It Up or DDR

QSC RAVE 88s Firmware Update failed, unit bricked (full recovery)

QSC RAVE Firmware Corruption via Update Failure

A while ago I picked up a pair of QSC RAVE 88S. These magic wonders from my favorite commercial audio company use technology from PEAK Audio/Cirrus Logic to shove audio across computer networks at low latency. This system is called CobraNet, and it’s implemented in hardware from a number of companies. I think it’s legacy now, with the newer Dante standard getting all the love. These RAVE 88S are pretty crusty, having been superseded by units from the QSC BASIS line. However, home user + budget = the cheap $20 units from eBay. I remember getting them working, then I put them on a shelf for later use. The 88S models each features 4 stereo pairs of digital audio in and out (AES/EBU, which is like the coaxial connection you find on some CD, DVD, DAT, MiniDisc and computer sound cards. 2 channels, not 5.1 or 7.1. It cannot do 8 channels from ADAT. AES/EBU is the professional version of the digital RCA jack, using a balanced XLR connector.) There are other models that are 16 in, or 16 out, and another series of analog in and out. I’d really like some of the analog ones like the 161S, but there is more competition to get hold of those.

MAGFest is coming up and this is motivation to get a lot of stuff working. I borrowed from my friend Robert 3 CobraNet endpoints from PRN. I didn’t have much luck with them but didn’t realize one of my RAVE 88S wasn’t transmitting any audio on the network so I need to go back and investigate further. The PRN units might attenuate the audio level based on room sound level — not sure. It’s in the patent.

Through the magic of looking at eBay, which I try to avoid, I picked up 3 QSC BASIS 904zz units. These are CobraNet endpoints, they’re much newer, and they have the functionality of controlling the amplifiers I own from QSC. I have an older unit called a CM16A which has similar functionality, but without CobraNet and without the DSP. I didn’t know what I was getting into, but the BASIS is just totally bad ass. Similar to modern DAW programs you can draw out “cable routing” between the CobraNet audio coming in, the DSP system and the amplifiers attached to the unit. Many band EQs, Crossovers, Delays based on feet per second of sound travel… doodle it in the software, sync it to the unit and it’s there. I didn’t even know it could do this stuff, I was just after CobraNet out that worked!

BUT… on to screwing up the firmware in my RAVE units. I was having tons of problems with the PAM units receiving audio from my RAVE units. The RAVE units are configured with this utility known as CobraNet Disco (CobraNet Discovery.) All the CobraNet stuff seems to be based off of reference designs, but I’m not sure 100%. After a lot of fighting with the RAVE boxes I finally decide that I should bring the RAVE units to the same version of CobraNet / firmware as the BASIS units are running and the PRN units are running. So I use the Disco utility + firmware snatched via (RAVE boxes are old!) and … clicked update.

The utility kicks in, does it’s thing till about 3/4ths of the way … then freezes. It says it can’t do it’s thing any more, and halts. I try again, no go. I try to upload the firmware that was in the unit originally, and it doesn’t work either. Take a deep breath, power cycle the thing…. and it’s a 1u rackmount brick.

I try setting the rotary encoders on the front to FFFF, no go. Apparently there is a recovery method outside the one I came up with, TFTP + RARP style that QSC has. But I don’t know the details.

I took my unit apart, because who doesn’t want to know whats inside. It’s a lot of screws. But once inside I notice after picking off a sticker that the main program store chip is an AMD flash device. I pull the chip, throw it in the programmer and save off the contents. For giggles, I upload the firmware straight from the HEX file that the Disco utility uses, slam it in the RAVE 88S and it boots fine. One caveat, and that is the MAC address is half FF’s. The vendor side is okay but the unique unit side is not.

In the earlier firmware I did find the MAC address near offset 00007C04 in the chip. On the earlier version of the firmware I believe the MAC address was in there straight, but the newer version has FF padding in between each octet. So you will need to randomize this otherwise the use of two units on the same network switch will go south.

The flash chip is an AM29F010B in my units.

Now, one of my units for some reason doesn’t work. It doesn’t take digital audio and put it on the network. I think it RX’s fine but just won’t transmit. This is what threw me off since I was trying to do use the PRN units with this one. The second RAVE 88S, I had to hit it to get it to work (no ethernet link/no LEDs at all except one pulsing.) Technical tap they call it. After pushing down on all the seated chips (probably GAL/PAL type chips since there are versions written on them?) I haven’t had an issue. I think the unit that won’t transmit audio — it might be a SNMP value that is wrong but I haven’t spent the time to try to compare the output of snmpwalk against both units (Disco utility I think mostly uses SNMP to do all control functions.)

I was also concerned the RAVE 88S wouldn’t lock up to 44.1KHz and would require 48KHz bitrate. This isn’t true, it has sample rate conversion internal. I’ve been driving mine with both a Denon CD player and a minidisc field recorder, single ended into the XLR.

I never did solve why Disco can’t update the firmware. I didn’t have it on an isolated network, so that might be one issue.

I do plan to try to make fan out cables from the DB-HD-15 connectors on the BASIS units to hook them to 3rd party self amplified PA speakers (the BASIS units I have are meant to only connect straight to QSC amplifiers via the dataport connector on the back, a VGA type cable (that has to have all wires in the cable separate and straight though! Cheap VGA cables not ideal – shared grounds!!))

Also, did I mention, the BASIS units are way cool!!

And in case you came here sniffing around, here is the last versions of the RAVE firmware (Note the topic of this post about bricking units trying to update firmware:
QSC RAVE Firmware Files – The last ones

Adaptive Microsystems ALPHA LED Sign on Lantronix MSS terminal server

Alpha LED Sign

Every few years I pull all this out again, and every few years I’ve forgotten a thing or two and have to poke around figuring things out again after loosing a part or two. MAGFest is soon approaching, and the goal is to ask attendees not to press start on the pinball machines more than one time.

Years ago I picked up some two line LED signs from Adaptive Microsystems. These were used in call centers, at least two from “Mac Warehouse” if anyone remembers that catalog. Probably ACD queue information provided by a middleware solution between the sign and a phone system.

These connect via serial line on a 6 pin RJ-11 (6P6C) cable. The business signs can do RS232 or RS485, while the consumer BetaBrite model can do RS232 only AFAIK.

I bought some Lantronix MSS10 units from eBay dirt cheap years ago when I originally bought these. I used velcro to hold the MSS10 boxes to the back of the signs and made cables to go from the RJ11 serial port to the Lantronix. I then wired in the 5vdc for the Lantronix straight from the power supply on the sign. This way there was only a single power cord and you could plug 10mbps ethernet straight into the sign and feed it data. All this was taken apart when these were located on Granby Street in a project for Art!Everywhere during the 757 Labs Hackerspace days. So I need to put it back together, and I couldn’t find the original cabling from when the term servers were on the signs just when they were hooked to a Livingston ortmaster for the art project.

As time has gone on a lot of pages have come up about the signs, much different than when I first got them. I recommend BB-XML for talking to the signs, it’s amazing. And Walt’s LED sign page.

The pinout to go from a RJ-11 6p6c shell to a DB25 female is:
GREEN to Pin 2
RED to Pin 3
BLUE to Pin 7
This is used with a rollover cable. This is where the two plugs facing each other tip to tip, the wires are the same on both sides (bottom is same color, top is same color.) See Walt’s page for info on this.

The pin full opposite of ground (WHITE) have a +5vdc line from the sign — it needs to be removed or protected from coming into contact with other pins. Insulate or cut the 3 spare pins.

In my old cables I swear I connected CTS to RTS on the host side or something, but I think I got around it this time.

The MSS10 setup should go like this:

Connect via null cable to the serial port:

Username: Whatever
Password: system
change ipaddress x.x.x.x
change subnet mask 255.x.x.0
change gateway x.x.x.x
change speed 9600
change charsize 8
change stopbits 1
change parity none
change flow control ctsrts
change modem control disabled
change signal check disabled

The MSS10 listens on TCP port 2001 :

my $moon = `uname -a`;
print "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\001" . "Z" . "00" . "\002" . "AA" . "\x1B" . " t" . "$moon";
print "\004";

./ | nc 2001