[weblacky]

For the most part my understanding is that any serious customers left on SGI's are basically running some sort of custom application solution that for one reason or another they're just isn't worth attempting to rewrite or they don't have the license to access the source code and it's a product that was dead-ended. We have also heard of some appliances or solution systems that used SGI's as part of their basic functionality that then to keep going requires you to keep the associated station going. We've heard talks of highly specialized printing and textile looms using Indy, O2, and Octanes.

I've heard of certain type of pick and place systems using a basic form of computer vision from the early 2000s that utilized the last SGI's as part of the appliance. When you're talking about something with $20,000 or $30,000 to replace at the same functionality it's still worth a couple thousand dollars to continually replacing parts of each SGI as it fails versus trying to replace the entire machine.

I've been led to believe that all the old simulators for flight or trucking or thereabouts have been replaced these days and that nobody is running the old simulators on Onyx with multi headed projectors for certifications anymore.


There's also old medical imaging machines that used SGI's for volumetric rendering of their scans into the various files and 3D representations/models. You don't see those much in the United States but a lot of those systems were re-sold on the secondary market and you hear about them overseas or in very remote hospitals that give basic MRI or PET scan ability even if the resolution isn't as sharp as today's they at least have something on site to help people. And to keep the certification and functionality they have to again repair the stations that come with the imaging appliance.

I have heard whispers of a few hush-hush entities that are still running cluster style systems with video processing that are meant to perform an assistive function of some kind that renders data from an older infrastructure or older form of public utility data.

Again most of it is economically based in that it's simply cheaper to keep it going. But a small sliver of it is either dead-ended product where there is no feasible replacement made by a company or entity at this time or it was a custom application and the source code has since been lost. Where it would have to be rewritten from the ground up and it's been determined that either while they're attempting or while they're thinking about that they're still running the old infrastructures.



In terms of capacitors I'm afraid you're a little bit mistaken based on the generation of machine. Indy mainboards for a fact had tantalum capacitors which have a pretty low failure rate although if you feed them bad or very very unstable power they have a tendency to sizzle and burn.

Indigo2 starts to show quite a bit of electrolytic caps, specially in the GIO64 back backplane module and gently sprinkled throughout the motherboard. Though again we haven't seen many failures lately but I expect to based on its age coming up and the inability to get rebuilt power supplies at the moment.

Now Octane and fuel and Tezro are all part of the next generation of circuitry. They all use a modern component process and the company started to use very advanced components for the day. One of the things they did on these three systems was used polymer capacitors and solid aluminum electrolytic capacitors. These are all capacitors that wouldn't normally leak and have an interesting benefit when compared to your common electrolytic wet capacitors.

However SGI never designed any power supplies. Every single power supply on every single SGI I've ever seen no matter the system was somebody else's design and normally carry lots and lots of electrolytic capacitors. To that end that's usually the primary failure point of any system to start with. Electrolytic capacitors aren't really designed to last more than about 12 to 15 years of constant use. They actually start to age faster when left on the shelf after a two year period according to online military research papers describing the problem of accelerated aging in storage. That's why rotating inventory and powering it on for a certain period to help keep the capacitors in good health is actually now a recommended procedure.

Of course that's not gonna magically heal something that's been in storage for 10 years. By then it may not even be safe to attempt the power on and a simply require a rebuild.

So that's the power supply is the first line of attack that needs to happen. A power supply giving bad power either in the form of ripple or the form of unstable voltage causes all the other parts in the system to work harder. They don't want to do that and so older components then start to fail out in catastrophic ways because the power you're feeding them Isn't really to their liking and they are no longer young enough to do that hard work to tolerate that kind of power anymore. Hence you have ceramic caps and things like that starting to go.

In the end Octane has the highest number of capacitors on the motherboard that I've ever seen in a desktop SGI. Yes they're not going to leak because they're not a liquid chemistry. So SGI did purposely choose the chemistry in attempt to stay off these types of failures But even polymer caps are not designed to be used for 20 years. They have a known lifespan because they tend to be worked pretty hard especially in power applications.


That's not to say that every capacitor in every place on an SGI motherboard has an equal chance of failure. But as I've recently been dealing with fuel motherboards and a few others they have voltage regulation module regions on their boards and some of them work very hard and so the failure profile is often the polymer caps in those regions go short and taking out the entire VRM, and sometimes a few things attached to it.


So while I would agree that SGI's are commonly very well made for their day and their price did reflect that. It's not the same as simpler z80 home computers from the 1980s where they could get away with simpler solid-state components at low power. A lot of times when you're dealing with lots of electrical current you tend to need a lot of filtering in it at various stages and electrolytic caps are the cheapest easiest way to perform that. That's just the nature of the beast.


Normally an old power supply tips the scale and takes out itself and a motherboard. Often times the motherboard survives but a certain percentage they don't and require component level repair.

They obviously made the power supplies all custom sizes until the very last line of SGI stations because they wanted to sell power supplies as part of their revenue stream. But now that they don't exist anymore you can't just pop down to the local store and have a hope of getting a power supply that's going to work in an octane, it's like a piece of medical equipment, it has its own power supply and its own constraints and now you have to deal with that.


The system might've been designed well but the power supplies weren't that much better other than being high-end and having as much intelligence as they could for the day they were built. They tend to have a lot of passives to create logic in the older power supplies including octane. Once you get to stuff that's a little more modern it starts looking like a power supply you would get into one of today's computers where everything is been consolidated into just a few chips With a bunch of regulation.

Every single SGI power supply has somebody else's name on it and they weren't designed by SGI.

I would assume if you're a collector that keeps their SGI on the shelf you should be buying a rebuilt power supply every 10 years or so. If you're someone that uses it for shall we say a few hours every quarter of the year then I would say you might actually be able to go 12 to 13 years with some confidence between intervals of buying rebuilt power supplies.


The motherboard capacitors are somewhat similar. But I would guess the same interval roughly.


The bigger issue on octane is the amount of heat. It's one of the hottest systems I know and that stresses the components quite a bit. My expectation of octane peripherals that use electrolytic caps would be definitely to have them replaced every decade and we're not really sure about octane graphics at the moment. It's been proven that the older graphics cards have an enhanced survivability compared to the Vpro line of graphics cards on octane.


We've seen a lot of broken solder joints due to thermal expansion as well as just having odd graphical glitches and problems that have yet been unexplained in a few publicized scenarios.


In the later model SGI's V12 failures are getting a lot more common and it's something I'm personally working on right now. That's only because there's a market for those cards right now. You don't see these failures as often with V10 even though they're basically the same card it tells you that the thermals are catching up with the design.


I don't make it a point to try to disassemble heat sinks off the chips. Most SGI's have some form of adhesive pad or adhesive compound holding the heat sink on the chip. It might be lightly holding it on where you can crack it off often times when you see people on YouTube try they end up pulling the entire processor out out of its socket if it socketed and still attached to the heat sink when they make the attempt.


SGI did not use much in the way of thermal paste. My feeling on that had to do with manufacturing tolerance and material longevity. You have to design a machine your surfaces a little more precisely to use some form of thermal paste to mate the two successfully. You'll find quite a number of these heat sinks are fairly rough and haven't been ground or lapped in anyway and so the tolerance is between the two surfaces are too gross and uneven to support thermal paste so they used thermal pads instead.


I'll also say that the thermal paste of the day was highly prone to breaking down over intense thermal cycling too. Some of these systems cycle so much heat that it's also possible that the common CPU thermal paste would just turn into a brittle chalky mess and stop doing its job after a few years with those kind of temperatures. Today we have a lot more options like ceramic paste and all that. I assumed it was a design and cost cutting measure and also meant that no one had to deal with improperly applied paste or any of it leaking or being tracked out and onto the board.

I've never personally heard of somebody attempting to use common thermal paste interface parts in an SGI.


Because it's not common to disassemble the heat sinks we don't have a lot of public knowledge on what size and composition all the fur pads are. Well obviously today you could probably use a standardize silicone thermal pad we would still need to know what thickness to use as most of the heat sinks are not spring-loaded to hold compression they're absolute distances.


there are some chips in the octane like the heart ASIC that I would not even touch with my hand as there have been reports of minor interaction with thermal expansion cracking solder joints to the chip and requiring a whole new midplane board.


SGI's were not designed to be stripped in the field. Those with a modular design like the octane were designed to have the entire module removed and then replaced by a field tech. Those machines that do not have modular boards were designed to just be part swapped in the field and the board taken back to SGI. I wouldn't call the boards well designed for the IT staff or even a curious owner to start removing components & heatsinks from. I would call that a bad idea.