Watts Up: Packaging to the End

Now that we have a working system, it’s time to package the new PSU so that it fits inside the SPARCstation-10. Given that the SS10 is a very slimline, cramped chassis, this took some ingenuity. I knew it would even before I started and had picked the PSU/components/mounting-location/etc cleverly to make internal mounting possible. But there’s a huige difference between theoretical “oh yes, that will fit in there” and the actual practicalities of allowing for screw-width, the mounting rubber grommets that the SS10 requires for drive bays, and so on and so on.

Craggy Island
In spite of not living on Craggy Island, it took four weeks to source a Dremmel-like mini hand-grinder/cutter/low-speed-drill tool. Hardware stores, ironmongers, friends and acquaintances all came up dry. *Eventually*, only yesterday (with RetroChallenge time running out), I found the last one in town in a petrol station, and snapped it up. Good job, too, because it would have been impossible to finish the prject without it.

Boxy Horror Picture Show
Two days, a complete weekend (apart from a break to play hockey) of measuring, marking, drilling, grinding, cutting, bolting, screwing, unscrewing, rescrewing… I’m exhausted, but got there in the end…

First up was fitting a fan (and the necessary cutouts and plastic bolts in each half of the plastic box enclosure obtained in Episode One)20150201_12044820150201_120521Oh yeah, and drilling the vent-holes. The fan blows “inwards”, from where the AC terminals will be to where the DC output harness and the vent-holes are – mandated by TDK NV1 application notes and handbook.

20150201_04240220150201_04242120150201_042444You will notice that the PSU is “raised” off the floor of the box by plastics bolts and a couple a “spacer” plastic nuts – necessary because the few surface-mount components on the underside of the PCB get slightly warm, so there needs to be at least some airgap there.

Final adjustment to the old APS-39 PSU chassis was a cutout to be able to route the incoming main cable so that it wasn’t in the way of the other cables:20150201_042527

Next up was a cutout in the main cross-bar of the SPARCstation-10 lid (originally intended to reduce RFI/EMI emissions from 1990-vintage disk-drives from effecting adjacent memory-modules, and vice versa, but not a problem these days, I won;’t be running a Seagate upside-down 10-platter 1st-generation thermonuclear Barracuda disk drive ever again!).
20150201_230044

And here is the completed adapted new PSU all wired up:20150201_195913

Finally, and this took several hours of sweat and tears, was to mount and install this monstrosity in the chassis, using the SPARCstation-10 drive-mounting grommets, which have to be very accurately positioned so that it will just slot into place. But after a lot of fiddling, measuring, drilling and screwing, I got it into shape to just slide in (and more to the point, just slide out again if needed)…20150201_225747.

And the main trick is to hide the cables underneath the diskette drive, just about the only “unused” space in the SPARCstation-10 chassis. The end result is pretty neat and tidy:20150201_225849

Conclusion
Yes, you *can* relieve yourself of having to limp along with the original APS-39 PSU in a SPARCstation-10, but it’s virtually impossible squishing anything else in there, especially something that can provide 5.1VDC @21A, unless it is a naked TDK Lambda NV1-350-TT or NV1-350-TT-N and some clever home-brew plastics (the pre-boxed versions of the NV1 are too long).

Finally, never assume that ATX crimpable female blanks and a standard AWG-18 ratchet-crimp-tool will go perfectly together. They won’t. You’ll usually need to pre-cut the tangs on the blanks so that they’ll crimp properly with the tool you have. And in spite of all the advice *not* to pre-solder the wires before crimping, you will probably have to pre-solder when doing two-into-one crimps.

Summary
It’s done and works, but it will be a long time before I talk myself into doing this again! I’m off for a well-deserved kip now…. zzzzzZZZZZZZZZZZ

Watts Up: Back in the Game

As discovered last episode, the SS10s are both still alive. Good ole serial console debug!

The problem running the SS10 off my adapted NV1 PSU is that the SS10 really wants 5.1V, not 5.0V. It will run rock-solid at 5.05V but anything below that is a bit hit-and-miss.

From Russia with Data
Although I had managed find numerous TDK Lambda NV1 handbooks, datasheets, application-notes and so on, from the UK and US TDK websites, they had no documents describing the really critical bits: does the PSU have built-in across-the-line AC filtering and line-to-ground filtering, and what is the actual maximum (trimmed) output voltage of channels 1 and 3?

As is so often the case when needing such specification details, the information was available from Russia. TDK’s Russian website, to be precise. Even though the “17159-20″ handbook is freely available from that site, I always feel a little bit surreptitious getting such information from tertiary manufacturers sites. A hang-over from kremvax days, I suppose.

Thanks to the Russian-sourced data, I can see that the NV1-175 (NV-350TT flavour) has user-adjustable trimming-potentiometers and the nominal 5V channel 1 is formally rated upto (and qualified for) 5.5V before the over-voltage protection cuts-in, here’s a close-up of the trimmers:20150130_165600

Twiddler On the Hoof
Most potentiometers are eg: 10-turn devices, but the suckers on the NV1 are way way way more than that, dozens of turns from bottom setting to top. I initially was turning the adjuster as slightly as possible to avoid overshooting (once you push the NV1 into over-voltage-protection, it’s a bit tricky to get it out again, except by blind adjustment whilst removing/reapplying main power).

Of course, my very delicate touches were making absolutely no difference to the measured voltage – were the pots broken or the Russian document incorrect, or what?

Poking a screwdriver into the guts of a live PSU is pretty scary, not something to take lightly, so big twists on the pots just didn’t seem like a good idea at the time. Eventually, after I plucked up the courage to twist the driver enough to make *some* difference, *any* difference, I managed to get the nominal 5V line to 5.1V (instead of the 4.98V it had been opriginally). However, without the PSU under load, the measured voltage wobbles around quite a lot, it just wouldn’t go stable. Given that I picked the NV1 particularly because it had no “minimum load” requirement, this was a bit disappointing.

After several hours without getting stability, flickering between 5.0V and 5.19V, I decided that I would have to make the tuning adjustment “in-situ”, actually hooked-up to the SPARCstation-10 with everything switched on. Now this is starting to sound very dangerous! If that big cap on the NV1 board blows or fails-short, well…

It still took a while, I had to pretend I had four hands (one for each probe, one for the jewellers screwdriver, and one for the mains plug). Eventually, though I could get a stable 5.1V on the nominal 5V lines and a stable 12.0V on the nominal 12V lines (which had been at 11.97V beforehand).

That should do it! For anyone else trying this: be prepared to take a *lot* of time here, these pots’ screws have some “slew” (changing twist direction does not take effect immediately).

Operating System
Now it was time to try installing an operating-system, Solaris 2.6 has modest requirements, and works very well on this class of machine, so that is what I chose. In this picture, the external CDROM drive and NV1 are connected, all ready for Solaris installation.20150130_114105. The stuck-on paper envelope is there to keep the fan-air over the CPUs rather than escaping into the room… touch-tests show that it is necessary… The O/S installation went through fine:20150130_112221.

The Moment of Not Telling Too Many Lies
Well, now that we have an O/S and a working system, surely it is time to try running the pair of 180MHz HyperSPARC CPUs that the original PSU could not manage?

It would be exactly that time, except that I have misplaced those darned CPU modules (Where’s Wilma?)!!!

Never mind, this gives me the opportunity to show the SPARCstation-10 doing something rather unusual, something that very, very few other systems of the time could do, and that many modern systems still cannot do: run three CPUs with different base speeds, cache sizes, and different number of levels of cache, all under control of a single operating-system instance: I have another single-CPU 180MHz module, and a dual-CPU 90MHz module. That all adds up to 360MHz, and that earlier dual-90 unit actually consumes more power and dissipates noticeably more heat than the newer die-shrunk 180MHz modules do, so in terms of stressing the new PSU, this configuration will be even *more* of a test. Does the system still boot? It sure does! (video). And repeatably consistently, yay!!!

Here are the views of the graphical login and desktop screens:20150130_153231
20150130_153747.

And the final piece of today’s cake: internet access using the ancient (and somewhat incapable) “hotjava” web-browser:20150130_155306How retro is *that*?

Watts Up: Episode 14 – Ear Cable

After suspecting that I had killed both my production and spare SPARCstation-10 motherboards, it is time to prove it one way or the other.

For SPARCstation-class machines, there is a handy trick that can be used to prove the main memory bus, the memory-controller, the CPUs, the basic “aliveness” of the on-board peripheral devices, and SBus plug-in cards: the serial console in firmware-debug mode. This spits a *lot* of self-test messages to the serial-port (at 9600 baud 8n1), even ultra-low-level tests of eg: “is the CPUs on-chip TLB storing, returning, and searching for entries correctly”, is the CPU-module cache-RAM working properly, are the hardware page-table-walkers working, etc. Extremely comprehensive, these identify any failing (or partially-failing) CPUs, DMA controllers, main memory MMUs, I/O-MMU, SBus devices, memory DIMMs, and so on.

Getting into Firmware-Debug Mode
The standard way to get the SS10 into firmware-debug mode: hold down the “Stop” and “D” keys on the Sun keyboard whilst powering up the system. Obviously, this won’t work if the keyboard is busted (or missing!) or if the keyboard UARTs or EBus on the motherboard are not working…

But there is another way to force firmware-debug mode – remove the NVRAM chip (the one with a barcode sticker on it, and mounted in a plastic DIP carrier) before powering-up.

Of course, to see the test/trace messages, you will need to connect a serial terminal (or terminal-emulator) to the SS10 serial-port via a “null-modem” cable. Although I used to have null-modem RS232 cables coming out of my ears, I have not needed one in anger for a few years and don’t seem to have any around any more; so I had to make one by rewiring a “straight-through” RS232 cable.

Trace Messages from Production Machine
Let’s see the results from the “production” SS10 (with HyperSPARC CPUs): (10-minute video). Hey, it is still alive, takes a while to come fully “up”, and can’t “see” the keyboard, but once it has booted fully, can access it over the network! Phew!

The problem of not “seeing” the keyboard is most likely due to the slightly iffy cable-socket on the keyboard itself: after being detached/reattached hundred of times over the last 20 years, the pins in the cable plug don’t always quite connect to the conductors in the socket. Takes a little bit of wiggling sometimes. Well, it’s either that or the secondary Zilog 8530 UART on the motherboard is busted, but given that everything else is working fine, I suspect that the problem is the keyboard socket.

Trace Messages from Spare Machine
To prove the death or otherwise of the spare machine (with SuperSPARC CPUs), I swapped-in the PSU from the production machine: (another 10-minute video). Hmmm, that one is alive as well, although it has a dead battery in the NVRAM chip (does not stop it from booting manually, just delays things whilst the firmware figures out that the NVRAM contents are corrupted). Phew again!

Of course, the next step is to see what we get when this spare SS10 is running off the new NV1-based adapted PSU… does *that* really work?

Watts Up: Episode 13 – Darning

From the last episode, the situation is that the old SPARCstation-10 fitted with a new NV1 PSU is not working any more. It had been, but now isn’t.

Swapping several parts from the dead SS10 into my “production” SS10 showed: the CPUs were good (*very* good), the TGX graphics card still works beautifully, and the memory modules are all still up to snuff; all individually and also all together.

Swapping the working (but limping) PSU from production into the non-working SS10 still did not bring it back to life: the keyboard beeps as expected, the fans turn, the HDD spins up, the floppy drive twitches, but after that, nothing. Nowt, nil, zip, nada, diddly-squat, zilch. At the SS10 end of the cable, the digital power-ready signal from the PSU was at 5.4V, a slight tad higher than I would ideally like – I’m pretty sure the SS10 would prefer this to be 5.0-5.1V ie: nominal-Vcc level.

Anyway, after what seemed like a lifetime laboriously retesting the voltages from the new PSU, at the SS10 end of the cable, and double-checking the wiring (pin X should go to pin Y – yes, tick!), the cabling and voltages were confirmed correct – I had not cross-wired a +12V line into a +5V line, or anything like that. No stray whiskers, crimps are OK.

Just in case, I waited a couple of hours to let the PTC (self-resetting “fuses”) on the motherboard reset, but still no joy.

At this point, the only obvious dead duck was the motherboard itself. Given that it is 23 years old (it is the original 501-1733 version of the SS10 motherboard), I foolishly convinced myself that it was just coincidence, it must have failed due to sheer old age…

Even more foolishly, I figured that the “fix” would be to swap-in the motherboard from my surviving SS10. Oops. That isn’t working either now, on either the old PSU or the new one.

Darn! Darn! Darn! So I now have two apparently dead SS10s. However, RetroChallenge 01/2015 isn’t quite over yet, I have a couple of (desparate) maneuvers to try, so… stay tuned!

Watts Up: Episode 12 – Jungle Drops

Sorry, but this is a longish multi-topic post, I have been a busy little beaver the last few days. Which only goes to show that busyness and productivity are not at all related…

Patchwork Revisited
My initial patch-wiring for the fan-power-distribution board had not been quite right – I also needed to bridge pin 4 of the backside connector to the positive terminal of the adjacent 3rd fan socket. It’s astonishing that simple “what’s connected to what” visual inspection can leave such a gap. Literally, gap.

Still, quite easy to fix:fanpatchAt least one would have thought so, but now only the 3rd fan would run. This took 2 whole days to get to the bottom of. It turned out that although my soldering to the pin was solid and good, the heat had separated the pin from the PCB trace through-hole via. Thus a resolder “and let it flow was needed. All three fans running!

The Law of the Jungle
As noted previously, I wanted to step-down the voltage driving the main chassis fans from 12V down eg: to the 7-7.5V that the original APS-39 PSU did. After quite a few calculations, for a triplet of parallel-wired 12V 0.22A DC fans, it seemed like a single 15-Ohm 2W resistor inline with the overall fan-power input would step the voltage down to 6.6V, which would be ideal – I don’t need the fans to rotate as fast as originally, because they do not have to cool the (now empty!) APS-39, just the CPUs and SBus cards. Said resistor would have to dissipate 1.9W, but that is within its’ spec… Needless to say, I had calculated (and sanity-checked, and double-checked) several dropping-resistor scenarios, for both the new PSU fan and the parallel chassis-fans:fandropcalc

However, I discovered the “law of the jungle” rule-of-thumb the hard way; with the above 2W resistor in series with the +12V input (in front of the parallel fan-out), it immediately ran *very* hot, too hot to touch. After a few seconds of running, I cut the power before the magic smoke escaped. With that amount of heat being dissipated, even via a large resistor body, there didn’t seem any way that the darned thing would survive for much more than a minute or two.

Quite a few ‘net suggestions were “always use a dropping resistor rated for *twice* the actual amount of power you will be putting through it”: this suggested that the 2W specification I had calculated according to Ohms’ law and the power law was, quite simply, only enough in theory, not *in practice*. Ouch! you might say.

Given a shortage of other high-power resistors and a shortage of time, the only maneuver available was to run a pair of these 15-Ohm 2W resistors in parallel (ie: 7.5-Ohms equivalent series resistance), which only drops the 12V source to 8.5V instead of 6.6V, but does reduce the power through each resistor – from 1.9W down to 1.6W
CIMG9529CIMG9530Not much of a reduction in power, only 0.3W, but it had an utterly drastic effect on temperature: now the resistors run for hours at just a tad over room temperature; by touch, you can only just tell that they are dissipating any heat at all. Amazing. The fans are a bit faster (louder) than originally, but it will have to do for now, until I can get my hands on a 5W 13- or 15-Ohm resistor.

And here is the dropping circuit attached to the fan-power distibution-board, all screwed-down in the APS-39 chassis, to keep it neat-and-tidy and out-of-the-way.20150125_152627

I know its’ a very minor achievement, but dammit I know you want a truly uninspired meaningless video, so here’s one for you: The Penultimate SS10-Chassis-Fans-on-NV1 Show.

PROM Night
The spare SPARCstation-10 that is recieving all this surgery would not be able to run the 180 MHz dual HyperSPARC CPUs for another reason than the original wilting PSU: the PROM (BIOS) chip is not new enough to support such CPUs. This was a deliberate thing, I had forgotten that I had down-revved the PROM chip to research old CPU compatibility for the Rough Guide to MBus Modules. It had a version 2.7 PROM chip installed, the very earliest version for the SPARCstation-10. Also, the embedded battery in the NVRAM chip (known as “CMOS RAM” in PC circles) had died, due to the system being in storage for so long.

Fortunately, I have a collection of SS10/SS20/HyperSTATION/SPARCplug PROM chips, all the way upto the very latest version, Ross’ 2.25R. Time to swap out the old PROM and NVRAM chips…CIMG9525That should be better:CIMG9526

The Intermittent Dead
So everything should now be peachy, huh? Well, no: sometimes the SS10 boots up OK, but sometimes it doesn’t. Nothing changes betweentimes. Measuring every voltage on every pin of the new PSU, at both the PSU end, and the motherboard end, shows that the voltages are pretty much where they should be: 4.98V, 11.99V, -12V (and the numbers are probably slightly higher – it’s a very cheap digital multimeter). Ideally, the nominal 5V lines should be at 5.05V (they may well be!). I double-checked the PSU sense lines’ level, they are matching the main power lines.

I double-checked the crimp terminations in both plugs, and replaced the couple of maybe-dodgy terminations. Still unpredictable: sometimes boots, but mostly doesn’t. Removing one PSU (still using the 60MHz SuperSPARCs at the moment) did not change matters, neither did disconnecting the fans (and the fan-voltage-dropper circuit). Heck, it was perfectly reliable just 6 days ago, how can it have gone so far downhill, and why? Keyboard beeps are still there, but most times nothing else is.

Damn. I’m getting a bit paranoid now – perhaps these CPUs have failed, or the graphics card, or the memory modules? Methinks it might be time for a swap-fest and prayers.

Watts Up: Heavy-Duty Fan-Holder

OK, time to fess-up: the 4th front socket on the fan-power distribution PCB is *not* the fan-power input: it actually originally provided a feedback voltage to the APS-39 internals, used by the APS-39 fan-speed control circuit – which is somewhat of a waste of time: anybody who has owned and run a SPARCstation-10 can tell you that the maximum spread of fan-speed adjustment (say between single 40MHz CPU and no drives, to dual-180MHz CPUs and two thermonuclear Segate “upside-down” Barracudas) is approximately 1/3 of a small hamsters fart.

Patch Wire Mod
The fan power input socket is actually a “hidden” socket on the back edge of the board, ie: it is enclosed in the APS-39 metal box, completely inaccessible – and we need an externall-accessible socket, so that the new PSU can be installed/removed without having to desolder anything – just plug-and-go.

Not to worry, the (extremely simple) distribution-board can be patched with a single soldered wire (a single conductor from a 10BaseT ethernet cable) from pins 1+2 of the back edge socket to the positive (red) pin of the front socket. That mod converts the 4th front socket into a usable fan-power input.20150122_002332
This mod preserves the logical wiring of the fans – parallel wired with inline staged voltage-dropping diodes:fanunwound2Those voltages are the ones measured from a running SS10 with it’s original APS-39 PSU. The fans are really 12V, 60mm, 0.22A; but as typical in 90’s computers are deliberately run under-voltage as full-speed is not needed and would be extremely noisy.

Reassembling the APS-39 Box
The fan-power distribution board fits into a recess on the inside base of the APS-39 case, and pokes the socket out of the side of the caseCIMG9514CIMG9513but now that the original internal glue-blobs, sticky-sponge and other crap in the APS-39 has been removed, this one remaining board needs to be insulated from the steel case – encasing the PCB in double-layer insulating tape is a bit over the top, but what the heck! CIMG9516.

Originally, this mini PCB was held in place by the bottom-half main PCB sitting on top of it. That larger PCB also acted as a “spacer” for the bolts that hold the sides of the case on. Now that I have deliberately destroyed and discarded that main PCB, I had to make a spacer which would also clamp the mini PCB in place. Soft plastic recovered from a bin (part of a discarded childs’ toy), and a little cutting and drilling, did the job.CIMG9517CIMG9521

Next up was to clip-and-screw the sides and top of the case together, with the mains cable for the new NV1 PSU routed through the middle, in place of all the original APS-39 circuit-boards, caps, coils, cables and crap; and finally to screw the fans back onto the side.CIMG9522CIMG9524
As you may have noticed, this was so easy I had time to make another cup of coffee, oops! It all almost works – only two of the fans run, so I suspect I might have disturbed the patch-wire when fitting it all back together. Also, the (two) fans are running at full 12V speed – OK for now, but I will need to replace the patch-wire with a 2W 13-Ohm (or 15-Ohm) resistor to drop the voltage down to approx 7V, but thats a job for later, I am waiting for delivery of suitable resistors from eBay…

For the main mains connection, the intent is to remove the 3-pin mains plug and attach the wires direct to the original APS-39 power-socket block (seen in the bottom picture – it fits into the end of the APS-39 case). The other small PCB in the picture is the first-stage AC->DC rectifier from the APS-39, so we’ll be losing that.

Watts Up: Fanny Thrice

Now that I have a system that initialises and passes power-on tests, I need to get an operating-system installed so I can test it under load. But first things first: to avoid the CPUs overheating, I need to get the fans installed and hooked-up.

The SPARCstation-10 fans were originally mounted on the outside of the old APS-39 PSU: this is by far the most convenient (or only) way to mount them properly anchored in the correct location, so I have preserved the now-empty APS-39 carcass to serve that purpose.

Although I mentioned previously that I had to destroy all the circuit-boards from the APS-39, that is not *quite* the case – I preserved the 12V fan-power distribution-strip from in-between the top and bottom halves of the APS-39 chassis:20150120_204952This is the only part just clipped in place, and thus is the only sub-board that can be preserved – fortunately, because this is the only board I need from the old PSU!20150120_222814and from the side, you can see where the fan-connector sockets poke out through cutouts:20150120_222623

There are three fans, and four 2-pole connectors to this distribution board – the 4th connector is the input (which is wired the opposite way around to the other three, but uses an identically-polarised plug, just to keep things confusingly simple!). The black twisted-pair cable is one from my junk-box, to replace the very short original red+white cable – I just needed to steal the connector off that red+white cable (by a stroke of luck, the internal crimped connectors on the black cable are exactly the same size, so they plug straight in to the other plastic shell without mods). Thats two crimps I have *not* had to do!

Next, there are a couple of repairs to make: after a lifetime of having the fans unplugged and replugged, tinker, tinker, the connectors on the board have come a bit loose, and the solder joints are breaking. First, superglue the plastic connector housings to the board then resolder the joints on the flip side.20150120_215407