SGI: Hardware

Hi.

I just hauled five lovely SGI computers home. An Onyx2 deskside, Indigo Elan, O2, indy and the reason for the topic: a Fuel.



I took it apart and removed some dust and other nasty stuff, it looks rather good now. However, after plugging the power chord into the machine I get nothing but a rather high pitched whine from the PSU and something I would describe as clicking but not quite, like metal changing shape due to heat. I get nothing from the L1, although my USB-to-rs232 adapter has been acting up.
After removing the power chord the whine persists for a while (I suspect some capacitors are discharging)

The PSU is a 460W hipro. Is this PSU worth trying to repair? I believe the fuel has a SGI specific powersupply which is rather expensive?

Edit: Spelling and formatting

Might be worth fixing, not something I'm up to doing myself. In general though, you might want to check out this thread on Fuel PSUs for general info and possible alternatives.

Thank you smj. I read up on the thread and others it referred to. I'll take a shot at opening the PSU and hope to find some broken components Meanwhile, I'll play a bit with the Deskside.

smj wrote: you might want to check out this thread on Fuel PSUs for general info and possible alternatives.

The ATX specs have allowed (but not required) PSU fan and power monitoring since 2.1. (see subsection 4.2): http://www.formfactors.org/developer/specs/atx2_1.pdf

There's a little more info on the function of those optional ATX power supply signaling here: http://www.freeopenbook.com/pc-hardware ... 6-TABLE-14

ATX Optional Power Supply Connector signals perform the following functions:

FanM

FanM is a 2-pulse/revolution signal generated by the power supply fan that notifies the system of current fan speed. If this signal drops, the motherboard realizes immediately that the power supply fan has failed, and can shut down the system in an orderly manner.

FanC

FanC is an optional signal generated by some motherboards to control fan speed for power supplies that are designed to allow this. The signal can range from 0VDC to +12VDC. A signal of +1V or less is recognized by the fan as an order to shut down, and a signal of +10.5V or more is recognized as an order to run at full speed. Intermediate voltage levels, which are supported by some motherboards and some fans, allow the system to instruct the fan to run at some intermediate speed. If this signal is left open (0VDC), properly designed fans run at full speed.

Sense

This is a supplementary +3.3VDC remote sense line, which allows the power supply to monitor the actual voltage at the motherboard connector of the nominal +3.3VDC rail and adjust it to stay within specifications.

and even more in this document: http://ww1.microchip.com/downloads/en/A ... 00764b.pdf Whether or not the Sparkle or any of the other Fuel power supplies provide (or use) any of these optional ATX signalling functions remains to be discovered.

The Fuel env, serial all, and hinv commands don't return anything related to the power supply that are part number specific (like some previous models have done), so there's at least a remote chance the optional ATX 2.1 PSU monitoring might provide the necessary info to the L1.

The strongest argument for an SGI-only power supply is jan-jaap's experience with the PS in his prototype Fuel becoming unrecognized after a firmware upgrade - which strongly suggests there's additional outside of the ATX-spec monitoring going on. But it might also be possible that j-j's pre-production prototype Fuel didn't include a PS with ATX2.1 fan and power monitoring capabilities. Monitoring capabilities that weren't expected by his original prototype firmware, but were expected by the later general-release firmware he flashed in. I only suggest this possibility because it seems improbable that Sparkle would have made two different power supplies with the same part number, one with I2C or similar circuitry, and another without this modification. If they'd made a one-off PS for SGI you think they'd have given a unique part number and raised the price accordingly.

Additional background reading - power supplies are discussed but I didn't catch a mention if any were the non-SGI Sparkle FSP460-60PFN (or included the optional ATX2.1 monitoring circuitry):
viewtopic.php?f=3&t=10849
viewtopic.php?f=4&t=14990&

So the unknown elements are how the L1/environmental monitoring circuitry in the Fuel obtains sense info from the PS, and whether the Sparkle FSP460-60PFN sold to SGI differs internally (beyond the MB connector pin-out) from the non-SGI Sparkle FSP460-60PFN sold to other vendors/the public.

Until that can be determined I wouldn't throw any serious money at a non-SGI ATX/EPS power supply. A Sparkle with the same model number as the one used in the Fuel (FSP460-60PFN) just sold on eBay for $9 - perhaps there'll be more with similar prices. http://cgi.ebay.com/460Watts-Power-Supp ... 0725423723

Thanks recondas, I'll do some research and take the PSU apart, if I find an i2c chip we'll know for sure. But is sounds more likely to me that it is ATX2.1. However like DELL have done, they might have swapped the pins around.. doesn't sound very likely though, but something to keep in mind.

recondas wrote: But it might also be possible that j-j's pre-production prototype Fuel didn't include a PS with ATX2.1 fan and power monitoring capabilities.

Fortunately, I still have that thing sitting under my desk

It's not a Sparkle. It's an NMB GM430WTXW01SSV rev. 02, OEM P/N SG002A430WSW (SGI P/N 060-0140-004 rev B)

jan-jaap wrote: It's not a Sparkle. It's an NMB GM430WTXW01SSV rev. 02, OEM P/N SG002A430WSW (SGI P/N 060-0140-004 rev B)

I have one of those, if you ever need a broken one

I googled a bit for the NMB PSU. It looks like it was also sold as the 'it002a430wsw' for the Tyan S2603 dual Xeon board .

It seems the it002a430wsw is a WTX PSU, a larger form factor for e.g. dual Xeon boards . Pinout of WTX is different but ATX-WTX adapter cables exist(ed).

Of course no guarantees that (1) the SG 002a430wsw is WTX and/or (2) that a reverse WTX-ATX cable would enable an ATX PSU to work in a Fuel.

jan-jaap wrote: I googled a bit for the NMB PSU. It looks like it was also sold as the 'it002a430wsw' for the Tyan S2603 dual Xeon board .

It seems the it002a430wsw is a WTX PSU, a larger form factor for e.g. dual Xeon boards . Pinout of WTX is different but ATX-WTX adapter cables exist(ed).

Of course no guarantees that (1) the SG 002a430wsw is WTX and/or (2) that a reverse WTX-ATX cable would enable an ATX PSU to work in a Fuel.

Thanks for posting that info jan-jaap.

I'm sure jan-jaap has already spotted it (or he wouldn't have included the pdfs), but the WTX specification also supports fan monitoring via FanC and FanM (see pages 23 and 29 of the WTX-Power-Supply-DG-10.pdf document provided by jan-jaap), and the data sheet he provided for the it002a430wsw specifically mentions connections for FanC and FanM functions - on pins 12 and 24 of the main 24-pin connector (though SGI might have opted for a different pin-out).

It might be interesting to see if there's any data/pin-out info available for the hipro power supply. Pontus could you post any Hipro model number info that appears on your PS (or attach a photo)?

There's one thing I can't make sense of: the WTX PSU spec calls for two large Molex connectors (P1 and P2), but so-called WTX form factor boards like the Tyan I mentioned, the HP XW6000 workstation (and the Fuel) all seem to use only one. This makes the pinouts in the WTX spec PDF rather useless.

WTX specifies the same 24 pin Molex connectors as E-ATX but the pinout obviously is not the same. The WTX PDF also specifies I2C pins (in connector P2) -- maybe that's where my blue Fuel's amazing 'no compatible PSU found' (or similar) message originates? The NMB unit doesn't have I2C. The only N/C wires documented are 4, 8 of P3. And they are N/C in the blue Fuel. How about other Fuels?

There's probably someone out there who figured out how to retrofit an (E-)ATX PSU in an HP XW6000. This might provide info on how to do the same on the Fuel. If the wire colors in the large Molex would be the same between the HP, Tyan and the Fuel, this would be another clue.

Is there a pinout available for the fuel power supply? If the extra pins are for fan monitoring/data lines will a fuel run without them? I'd go as far as making a small adapter for normal ATX supplies if there is a market.

astouffer wrote: If the extra pins are for fan monitoring/data lines will a fuel run without them? I'd go as far as making a small adapter for normal ATX supplies if there is a market.

Thanks for the offer, but based on some of the nekochan discussions (linked above) on the use of standard ATX power supplies, the answer seems to be the Fuel won't run (or even attempt to power up) without an active connection that reports the power supply fan speed/status (and possibly other bits of feedback from the PS).

If we can discover a working alternative Fuel power supply, a wiring adapter might still be a very viable (and needed) option, so please stay tuned in.

Can anyone provide a detailed photo of a fuel power supply board? I hope they don't use that encrypted ID chip like the motherboard does. If all the supply does is query a chip for an ID string or look for some fan signals I could program a PIC chip to fake those.

Hi all

I took my PSU apart today and can report the following. It is labeld HP-W460GC33 and has SGI stickers with 060-0140-007 REV:B and SGIW460GC33. The circuit boards on the inside is also labled SGIW460GC33 so it was obviously made on order from SGI. There is also SGI-W460GC31 on the primary side circuit board as well as IT-W460GC31 on the secondary. Probably the circuit board was use on several models. Googling w460GC31 gives pictures of the insides of an ATX PSU that looks very similar.

I checked all IC's I can find and didn't any i2c circuit in there. There is one Weltrend wt7517 which is a PC POWER SUPPLY SUPERVISOR with over voltage/current detectors as well as a FAN C function.

Checking the pinout on the 24 pin P1 connector it doesn't match the color coding of neither ATX or WTX specs, but the _alternate_ pinout here http://pinouts.ru/Power/wtxmain_pinout.shtml matches except for one pin. That page also states that there is a 6-pin header used with the alternate pinout for CPU power, which the Fuel uses (well, it's a 8 pin connector but 2 are not connected)

The fan header goes directly to pin 12 and 24 in the P1, the silk screen on the circuit board on the other end of the cables are labled FANM1 and FANC1. FANM1 is directly connected to a pin on the fan header. I'm guessing FANC1 goes to the wt7517 and the fan header by the looks of it.

So, I believe that there is no i2c shenanigans and that this is a WTX PSU with the alternate pinout and the fan is what the Fuel L1 is looking for. This could perhaps be verified with a working PSU and someone who can measure the voltages.

Thanks for pointing me in the WTX direction.

I also found a dead bug and some goo inside. I'm hoping some cleaning of the PCB in question will do some good.

Here are some pictures I took.

Somebody has serviced this before.

The WTX standard looks pretty dead, even the website for it no longer works. The additional connector of the WTX spec lists optional I2C data and clock lines. You guys are gonna make me buy a fuel now to solve this problem

Doing some searching it looks like the fuel talks I2C to other addons like the video card and you can even do some debugging from the prom. My first thought was maybe the power supply has the same monitor chip as the motherboard and the prom makes sure they agree before proceeding. The bottom line is someone with an oscilloscope needs to see whats going on with all these unknown lines.

astouffer wrote: You guys are gonna make me buy a fuel now to solve this problem

I'm pretty sure we could arrange something ...

I haven't given up on this. I did however break my hand which slows down most things. I've decided to try and convert an ATX PSU while trying to get the original fixed.

Ouch! Not good. Get well soon!

Some progress!!

I bought a decent ATX PSU and extensions for the 24 pin P2 and 8 pin P3 motherboard connectors this way I could easily move the pins around and add FAN signals to P2. I hooked things up in a mess of wires on my machine room floor. First gave power to the PSU with just my adapter to see that there were no shorts. Next I connected P2 to my Fuel, this way the 5V stand by power is available and I can try to communicate with the L1 port.

Hooking up my terminal and trying different baud rates I only get garbage. But very consistent garbage, hmmmm. I get a bit confident and try to power the beast up, nothing but more garbage on the terminal. I get more confident and hook up P3 (8 pin 12 V supply) and press power again. Now I get a beep, slow spinning fans and a red light that turns into a steady white blinking and more garbage on L1.

No bad! about here I figure out that the baud rate should be 38400 which my old vt320 can't handle. Time for an upgrade I now hook up my laptop and connect with the correct baud rate and voila! L1!



This is where I'm at. I'm going to play some more with this now.

And thanks for you're concern tingo but don't worry, I'll be able to get you the ND-110 safely to you anyway

Outstanding!!

Very nicely done Pontus - Fuel owners worldwide you you a debt of gratitude.