[weblacky]

Hi All,
   I used to be a Nekochan lurker a few (9) years ago...life got busy...now things are what they are.  I own several SGI stations and lots of dev software.  Those of you from Nekochan may remember my Tezro issue (placed too many PCI cards (filled it) and found with a Firewire card installed I had a VRM module kernel panic within 10 minutes of booting.  Replaced the VRM, no change, removed the Firewire PCI card...problem went away...that was me :-)

I've shaved down my SGI collection to the basic best: Indy R5000, O2 R5000 (@ 200Mhz but have a 300Mhz CPU somewhere...I bought it, but never installed it), an Indigo2 Hgh-Impact, and a Dual 800Mhz, dual headed, Tezro that I arranged to buy over Nekochan around 2010.

OK, now that I've introduced myself...I feel like I need to express a topic no one seems to talk about...but I was hoping to pick some brains,  Preventative maintenance of SGI Power supplies!

I've left my units unplugged for years now, I've not started them...nor really had the time right now.  But I'm worried about starting them in the future.  After all...they worked when I shut them down...but that's no guarantee for the future.

So what do people think about getting together a list or at least sharing some experiences of what actually "went" in their PSUs (all SGI models) and coming up with a list or plan to "refresh" PSUs...that haven't yet failed.  None of these PSU are new, period...we all know the PSU is the saving grace to preventing board damage due to out of spec power.  You don't want to power an out-of-spec component!  Replacing before power-on and you avoid death.

I have pretty good soldering skills (professional station and micro tips as well, but no hot air or PCB board heater) and am not afraid of removing and replacing basic through-hole components or even larger surface mount components (including ICs).  I own an ESR meter and a ring-inductance tester. 

I'm more worried about my Tezro than anything...since it's the most complex unit I own.  But I won't leave anyone out if you have valuable info to say about your PSU repair experience or the like.

Before they fail, what components should just be whole-hog replaced (due to heating, aging, duty cycle, etc..) to prevent destruction of this vintage equipment?  I understand there is the adage of "if it ain't broke, don't fix it".  But judging from stories, I'm unsure SGI designed these stations to survive a PSU component failure (that doesn't blow the fuse).  We have no support and no supply lines, so I think we should throw that adage away and instead think of something like, "put fresh gas in it".  We could just say, replace ALL caps.  OK, but what else?  All Diodes? Thoughts?

Also I realize I'm not differentiating between hot and cold PSU sides where issues change.  On hot sides it seems filtering caps, overheated diodes, and over-volted transistors seem common for consumer appliances.  But on the Cold side, you have Logic and grounding caps, perhaps just dry solder joints.    I'm looking for insights here.

Thanks for sharing any input and photos you all have!

[Raion]

Several of our members have discussed this in the past. The big thing is that the older a PSU gets, the more likely it'll fail, especially if regularly power cycled.

Some of our members have designed adapters for more modern power supplies or have suggested that replacment boards using modern off the shelf components could be designed.

I lack the skills to solder and don't want to deal with it, personally

[weblacky]

Unfortunately your suggestion is not applicable to the vast majority of SGI stations. I'm specifically looking for people who have changed components in the OEM power supplies. Given the space constraint of several of the stations there never will be a way to adapt a normal ATX sized power supply to them. If someone ever makes a 500w or 600W micro or Pico ATX supply then perhaps so.

Unless some magical, retired SGI employee leaks schematics or other technical documents online the stations are unfortunately too complex to reverse engineer easily.

The stations cannot be lumped in with Commodores or amigas. They are orders of magnitude more dense. Keeping the original hardware going is pretty much the only thing we can do. Due to the fact that there are no emulators for the stations nor any third-party companies currently supporting them, we're left to your own devices with time being against us.

Also I have no idea why anyone would be against this. These are not magical items. Any PCB can be filled with components from other sources. As long as custom programmable micro chips are not damaged you could theoretically replace every other component on the board with its expected replacement. Perhaps that's one way to go about it. But once the board has experienced a failure you might blow a trace or cook the board in a way that ruins the underlying circuit. If we just had clean boards were you had to solder all the components to it it wouldn't really be much different. The only exception being the tons of logic that appears to be present in the late 90s and higher SGI station PSUs. Where the power supplies are not just dumb power supplies they have their own tiny surface mount logic components as daughter boards around the power supply.

[Raion]

Easy easy, no need to get hostile or argumentative. You assume I don't know how complex these machines are. I did found this website, after all, and have owned many of them including the Tezro and others, so it's safe to assume I know as much as you.

The Fuel and Indigo2 are two machines that have had their respective PSU's reverse engineered to allow modern ATX PSUs, however, I agree with you this won't work for a significant portion of the range for a number of reasons:

1. Modern ATX PSUs don't fit
2. They can't always provide the high voltage differentials necessary for onboard components
3. It just is plain wrong.

A user here, however, has started looking into reverse engineering the SGI Indy and a few other PSUs and possibly seeing if a new PSU can be made from off the shelf components. That's what I meant by reverse engineering.

I'm not against it, either. I just said I lack the skills to do it myself and would rather not bother with trying to fix these myself. Each time one has failed or shows signs of failure, I send it off for repairs to someone else here. That's how I do it, at least. That being said, this topic really interests me.

[jan-jaap]

One of the main issues when replacing an original SGI PSU with something adapted from a PC ATX PSU is that the SGI PSU supplies a lot of current on the 5V or 3V3 rail, and the rest is mostly aux power for hard disks etc. Modern ATX PSUs supply the bulk of power on the 12V rail.

I own many more SGIs than I use on a regular basis, so often a lot of time passes between usage of any system. I refuse to leave them connected; two dozen or so 20 ~ 30 year old systems in a house where I live with my family is a fire risk I'm not willing to take. And that's before we talk about standby current.

This is my own experience:

• Personal IRIS: has a reputation for dying. Simple, accessible design. May consider proactive replacement of electrolytes. People have modded AT (not ATX) PSUs into them.
  
• Indigo: never had any issue other than the time I tried to use an R3K PSU in a loaded R4400 Elan system -- poof.
  
• Indigo2 "teal" : seems to keep working
  
• Indigo2 IMPACT: if it has problems starting up / switches off unexpectedly: replace 4 caps inside: http://www.sgidepot.co.uk/i2psu.html In my experience, success rate is 50:50
  
• Indy "Nidec": bad reputation, don't know specifics.
  
• Indy "Sony": have at least three, they never failed.
  
• O2: don't seem to have specific problems I'm aware of.
  
• Octane "lucent": has a bad rep, I never had one I think.
  
• Octane "cherokee": never lost one, but have a dead some where. Couldn't figure it out.
  
• Fuel "NMB": seems prone to dying. So far mine didn't. ATX adapter exist.
  
• Fuel "newer type": seems more reliable.
  
• Tezro: I think it uses a generic ATX power supply. You can find the generic type on eBay.
  

Bigger Iron:

• Professional IRIS: Blue L&H PSU. Mine seems to have developed a problem with the main caps, but they're expensive to replace so at least for the time being I replaced the entire PSU with one from a PowerSeries.
  
• PowerSeries: PowerOne SPM5 modular PSU. I've seen more than my fair share of the internals of these, I'm afraid. Mostly the PSU backplane seems to die and it's not a simple matter of replacing caps. Possible problems with the aux/bias inverter and optocoupler. I went so far as to partially reverse engineer the schematics, but to be honest: my most successful way to repair these was to buy a suitable SPM5 chassis on eBay and transfer the SGI specific modules into it. The beauty of generic, non-proprietary designs
  
• Crimson: if you've got a Cherokee PSU ("black") you're out of luck. They seem to self destruct if (when) they die. I replaced mine with the PowerOne ("silver") model. This requires a different 5V bus bar so is not a simple part swap.
  
• Onyx/Challenge: power systems are complex and don't age well. First there's one (or 2 or 3) OLS to convert 220VAC to 48VDC. All but the last rev are a fire hazard and should have been recalled/switched out except they more often than not aren't. Then every board in the system converts 48VDC to whatever it needs using DC-DC converter modules. These are sealed and not serviceable. The systems boards are so thick I've never been able to desolder anything from them.
  
• Onyx2/O2K: single, large PSU again. Death symptoms: doesn't switch on at some point. The insides are hostile, it puts out screwdriver-melting amounts of current and they are controlled by the MSC in an undocumented way. I've never been able to power on up on a bench, outside the system.
  
• O3x0/O3K: goes back to the Onyx design of an OLS and local DC-DC converters. The local converters are replaceable though, not soldered in. The PSU of the O350 has a design flaw documented by Intel (the maker) where a PFC diode ages and blows up. If you replace it it comes back to life.
  

But general, if (when?) a PSU dies it simply stops working. It doesn't take the system out. That doesn't mean repairing the PSU itself is easy though. But I don't really believe in a blind swap of e.g. capacitors. I try to have spare parts, and take it as it comes. I fear mostly for the big iron. Big, proprietary PSUs are a pain to repair. Big currents means significant risk of catastrophic damage if something fails.

Back to the Tezro you're worried about: that would be the least of my worries. I'm pretty sure any half decent ATX PSU can be used in it.

[weblacky]

Hi Jan-Jaap,
  Your name seems familiar to me, I'm sure I ran across your postings more than a few times, years ago.  First off, thanks so much for sharing!

Now that is some interesting info,  I'd not heard that Indigo2 PSUs had any successful conversions.

Do you actually have information on bench starting any of these Power supplies?


I've never come across that topic before, so if you know what to jump and where to places loads (and by how much), that would be a big help.  I've been looking into finding cheap PSUs to experiment with (like Indy PSUs) and such for now. If I could bench start and voltage check them before installing them...that would be great!

I've been reading up and looking into PSU rebuilds lately, actually I feel this work purposely has been given an unfair wrap...maybe to ensure people don't figure it out (remains a high-cost service).   I know it has high voltages...but I'm unconvinced at this time that you cannot, simply by checking (using a variety of methods) the, resistors, diodes, caps, and possibly current leaking transistor-like components to ensure they aren't (or have not already) fallen out of their respective ranges and that you couldn't prevent PSU issues or deaths by doing these replacements.  Especially in the hot-side of a PSU, the area that takes the most heat and stress.

If I held in my hand a high ESR cap, or a leaking diode from my working PSU..wouldn't replacing them be the healthiest for the PSU?  From what I've seen, overdriving semiconductors due to out of spec caps and resistors is the big cause of semiconductor failure, and that given fresh basic components, low-voltage semiconductors don't have a fixed life span (outside of environmental factors).  Semiconductors that DO bare a brunt (MOSFETS, etc) do need replacement after time...they work hard.

I'm also considering coming at this from an analog signature analysis viewpoint, given a known-good PSU and it's duplicate from the same manufacturer (unless there where production upgrades) they should be identical in component layout and values.  Obviously each manufactured variant needs to be investigated using this technique...but it's a real possibility.

I think after the PSU has gone "pop", that's another issue.  But if the fuse is intact and no pop has yet been observed, can someone come up a cogent argument as to why this mind set wouldn't be possible to successfully extend a PSUs life another 15 years?  Then doing the same thing again.


In my studying of PSU repair, I've discovered a duality to it. There are those (rare) individuals that have the insight to figure these things out (to reverse engineer enough to generate schematics) and those that can cheat and find the components that have failed/going and simply replace them with equivalent stock.  I'm more of an advocate of the latter.  I'm not looking to "improve" the designs...they are what they are.  But instead of waiting for age-related failure, can we take that option off the table for components we KNOW to not live that long.

I think we could all say that most electrolytic caps (outside of exoctic materials) were never made to stay in spec for 20 years (even on the shelf, most ratings say they slip even in new old stock) and many resistors may increase resistance with age.  It's a known thing.  Stressed diodes will turn into shorts easily.

So I'd be more of a fan of doing a whole-hog replacement on MOST PSU caps (I might do testing on the large filters, and if they are still perfectly in spec...Id' be inclined to leave them for the time being), but the rest...go now.  I'd then look into any diodes for leakage (breakdown voltage issues), then check all the resistor behaviors (IV Curve) then recheck values on current path resistors (at least).  Replace the rectifier or power diodes and the high-speed transistors (TRAICs or whatever they are) without question.  The film caps I'd check before replacement, and I'd quick IV Curve check the surface mount caps for shorts.  After that...I don't know of anything to expect a problem with?

I think when I mention this topic, most people see one issue, if you going to do that much work...why not wait until it explodes...same work right?  I'd proposing that it's not the same.  Once breakage occurs a more skilled technician and diagnostic equipment is needed and possibly physical repair of the traces and replacement of hard to find ICs is needed (read: expensive leading into unfeasible).  

My idea is refreshing working units (possibly repair bad units that still have intact fuses) and is why I asked about failure points and as well as people's openness to the idea.  It's a little expensive and if you're going to do the work yourself...it may take a few sittings, but it's often much cheaper than finding a working PSU (which has all the same issues).  I've not heard of anyone even offering working SGI PSUs that were rebuilt in any way.

I think there are two groups of collectors (and I've hovered between these groups myself).  Those that want to hang the cases on a wall as art, to be seen and admired, to think that the 90's was so much better than now and such...and those that want to collect the items and play with them sometimes and have full confidence that the value is both in the cosmetic condition as well as it's functional condition.

Do I plan to boot my SGIs often...no.  But I'd like the confidence to know they will work and I can start them safely at 3-5 year intervals at least.

I fear that, as wall art, when I feel like booting my SGIs and changing out some add-in card I got on eBay or playing with the old compiler for some reason...will they actually start and my collection won't go up in smoke.

I think being a vintage computing collector means you're the custodian of the item and it's your job to care for it until the next owner.  I don't like the idea of me saying: "It worked the last time I booted it...many years ago".  Yes, there are unknowns, the PSUs, don't need to be one of them.

If anyone has PSU info, please consider sharing it now, cheap SGI PSUs are getting rare and given time...all inventory just ages out.  If we had a recommended "parts list" for collectors to specific PSU/manufacturer/model units ,people could decide for themselves if they want to wait for a failure or if that SGI they JUST bought and seems to work now, should be rebuilt right now.


Thanks for sharing your info and stories surrounding these amazing machines!

[jan-jaap]

Hi Jan-Jaap,
  Your name seems familiar to me, I'm sure I ran across your postings more than a few times, years ago.  First off, thanks so much for sharing!

Now that is some interesting info,  I'd not heard that Indigo2 PSUs had any successful conversions.

Do you actually have information on bench starting any of these Power supplies?

The Indigo2 PSU is not a conversion but a repair / blind replacement of parts (electrolytes) that apparently often die.

The only PSU I know  how to start on a bench is the PowerOne PSU from the PowerSeries. Back when these things first started to fail on me, PowerOne still existed as a company, and I downloaded all relevant information from their site. The modular nature of the PSU makes it somewhat easier to troubleshoot, and the fact that you can still buy parts on eBay doesn't hurt either (they sold in many markets, mostly industrial and medical. Teradyne testers have them for example).

I've been reading up and looking into PSU rebuilds lately, actually I feel this work purposely has been given an unfair wrap...maybe to ensure people don't figure it out (remains a high-cost service).   I know it has high voltages...but I'm unconvinced at this time that you cannot, simply by checking (using a variety of methods) the, resistors, diodes, caps, and possibly current leaking transistor-like components to ensure they aren't (or have not already) fallen out of their respective ranges and that you couldn't prevent PSU issues or deaths by doing these replacements.  Especially in the hot-side of a PSU, the area that takes the most heat and stress.

If I held in my hand a high ESR cap, or a leaking diode from my working PSU..wouldn't replacing them be the healthiest for the PSU?  From what I've seen, overdriving semiconductors due to out of spec caps and resistors is the big cause of semiconductor failure, and that given fresh basic components, low-voltage semiconductors don't have a fixed life span (outside of environmental factors).  Semiconductors that DO bare a brunt (MOSFETS, etc) do need replacement after time...they work hard.

True in theory, but real life dictates otherwise. Time I can spend on this hobby is limited. In 2012 I had an annex to my house constructed for my SGI collection and I still haven't put all the wiring in. Like many others, I have piles and piles of parts in storage, without knowing what is where and whether it even works. I need to reduce this to a set of known-good spare parts relevant to my collection and get rid of everything else. It all costs a lot of time and takes priority over pro-active maintenance. Over the years I have also learnt the hard way not to mess with things that are not broken (not the PSUs but more the rest of the systems, especially the big iron)

I think we could all say that most electrolytic caps (outside of exoctic materials) were never made to stay in spec for 20 years (even on the shelf, most ratings say they slip even in new old stock) and many resistors may increase resistance with age.  It's a known thing.  Stressed diodes will turn into shorts easily.

This isn't incorrect, but IMHO things are a bit more complicated than that. First of all, the same logic applies to every other part of the system. Do you intend to replace components in the rest of the system as well? Second, there are many components, even in a PSU, which are not so easy to replace. There may be ICs, either undocumented or no longer manufactured. In one of my PowerOne PSUs it was a UC3844N IC that failed or went out of spec, not the passive components. I tried to find a pin-compatible replacement for an optocoupler once and came up emtpy handed.

SGI systems are not immune to other breakdowns, often heat/stress related. Just look at an RM4 from a Reality Engine: dozens of large QFP chips on a fairly flexible PCB. Throw some brittle, aging soldering joints in the mix and you know it's not going to have a happy end, but impossible to repair without expensive equipment like a large reflow oven. Worse for BGAs, though less likely to fail.

My idea is refreshing working units (possibly repair bad units that still have intact fuses) and is why I asked about failure points and as well as people's openness to the idea.  It's a little expensive and if you're going to do the work yourself...it may take a few sittings, but it's often much cheaper than finding a working PSU (which has all the same issues).  I've not heard of anyone even offering working SGI PSUs that were rebuilt in any way.

Ah, here's one for Indigo1: https://www.ebay.com/itm/PSU-REFURBISH-A...3574226040

PSU repair shops exist, but they don't have the SGI systems so they can't test anything and work more or less "in the dark".

Look up the series about the Xerox Alto restoration on Ken Shirriff's blog: http://www.righto.com/  These guys did a great job, a true inspiration. You'll also see the extraordinary amount of effort, time and knowledge it takes. Plus documentation, which is sadly lacking in the case of SGI systems. But despite the extreme rarity of the Alto system, I think they just replaced broken parts. NB: I have a similar L&H Research PSU in the  Professional IRIS I managed to get back up and running recently.

[weblacky]

To quickly address this...I had actually intended to replace internal main board components when they fail (on my desktop stations anyway). However looking into these, I don't expect much to fail in the next 15 -20 years outside caps. The stations are dense with ICs...but from my perspective they should be feed nice, clean, power - well within their tolerance. While random things do happen, I've personally not seen a logic IC go without one of it's dependent external components malfunctioning (IC was forced to sink too much current on a leg or something due to failure of other components...but your point is well taken...now you have more dead components that need replacing).

I was hoping that limiting these obvious failures would mitigate the type of IC failure you spoke of. Seems a lot of designs use simple method of pulling up and down IC legs and many of those techniques can cause damage if the neighboring components fall outside their original values. It's that situation that I see as our opportunity.

I've got equipment on order right now, hopefully as I get further into this, I may have something.

But, I do see your point on the "things work and life gets away from you" ...I'm only recently re-entering the SGI community after an 8 year gap...so I get that. I guess for me, I was considering making a project out of "refreshing" each of my SGI systems (new batteries, new caps, PSU check, SATA SSD -> SCSI Adapter retrofit, possibly new fans (if old fans are a problem now). Perhaps a stock OS reload on a few of them.

Then I'd feel better for the next 10 years.

[Trippynet]

I had one of the PSUs from my Indigo2 professionally re-capped a couple of months ago. It had started intermittently powering off, and a firm in the UK specialising in PSU repairs agreed to re-cap it and fix the random power-off issue.

Unfortunately after testing it a few weeks ago, it still has the power-down issue - although the re-cap job on it looks professionally done. As the firm in question provides a 1 year warranty on their work, I sent it back to them (no extra cost to me), it's due back tomorrow so I'll have to swap it in and do some testing to see if it's now fixed.

Of course, using a proper firm to do this instead of a hobbyist repair means it's not the cheapest approach (cost me just over £100). But if it means I ultimately get a working PSU with hopefully many more years of life in it, I'll be happy enough.

[weblacky]

Hi Trippynet,
That's good information, after you get it back (and have confirmed you're satisfied it's working) could you enquire as to what they had to look at to fix the issue when you sent it back to them? Did they not replace ALL the electrolytic caps (just ones that looked bad..I've heard that one before) and did they actually find something else and replace it, a burnt resistor perhaps? If you could please follow-up with them as to what they touched (even if they can't tell you the exact part location) what kind of part was it?

Hopefully they will say, some part name, as its owner you had to send it back, which means I assume you paid shipping again. Under that guise I'd say you're entitled to ask what happened so in the future after the warranty expires you know what was fixed last time.

That could give the community a valuable insight into a failure point.

Thanks for sharing.