The collected works of jan-jaap - Page 21

Next up, I loaded all boards of the 4D/70, measuring the power consumption (from the wall) each time a board was added.

With all boards installed, power consumption hovers around 400W @ 230VAC. Which means a little over 300W (from the wall) goes to the 5V rail, worst case scenario. Assuming 80% PSU efficiency, some 250W (50A) is a more realistic number. In either case, it is well below my 150A budget The PSU does not make any nasty noises like the original LH Research PSU did.

This is what it looks like right now:

Don't worry about the pixel faults on the screen, it's the LCD which is faulty.

Something else I didn't expect: airflow is top-down, like an Onyx2 deskside. The exhaust is at the bottom of the sides of the main tower. So my plan to install a thermal kill switch above the card cage will have to be re-thought.

I'm quite happy with my progress so far. I've solved two of the problems of this system: a faulty PSU and a faulty blower. I will still have to deal with a sick GM1 board and the system disk doesn't spin up.

Now that I've established that (1) the original PSU was faulty, and (2) the Power-One PSU can handle the system, I'll redo the wiring harness to the PSU properly and drill some holes in the chassis to mount the PSU.

If you're going to replace capacitors inside a PSU, beware that there are more parameters to electrolytic caps than capacity and voltage. In an SMPS you should use low ESR elcos.

Another important parameter is the maximum operating temperature. Higher max means it will last longer (at moderate temperatures). Use 105deg, 80deg is budget crap.

I've done my fair share of power supply repairs, and I prefer Panasonic FR series. FM series is probably OK also. You can buy them from Farnell / Element14, Digikey, Mouser etc. eBay is more for desperate stuff (components not manufactured for a decade with no suitable replacements available etc).

Spend 2.5K (plus dual 220V/32A service) to emulate a subset of apps for one obsolete OS on another obsolete architecture? Right ...

For hobbyist purposes, a Prism is much more practical. This is a HPC rig.

ivelegacy wrote: Before this idea, I was planning a diesel generator engine to power feed all my gears.

Don't know about Italy, but where I live you will never ever get a permit for that. If you live within city limits at least.

duck wrote: The random disk I have in my Octane just clocked in some 85MB/s in sequential reads off the disk

The internal SCSI bus of an Octane is UltraWide SCSI, 40MB/s. Faster controllers like the QL12160 work in the Octane, but you need a PCI cage/shoehorn. The PCI-XIO bridge used there is your next bottleneck because its limited to something like 150 ... 170MB/s throughput IIRC. In the case of the shoebox this bandwidth is shared by the three 64bit, 33MHz PCI slots. So to get sustained gigabit file serving performance you would need two XIO-PCI bridges (a shoehorn plus a second shoebox /shoehorn for the gigabit ethernet card). You also need an external enclosure for the disk(s).

It's not impossible, but an Origin350 is a much more convenient platform for this. Plus you can install a SATA controller in it, so you can get a half decent amount of storage without the need for one or more shelves full of crappy old SCSI/FC disks.

Never turned on? Are we looking at the same Origin350 or are you being sarcastic?

It's missing a drive bay door, most of the 19" rackmount rail kit, and the NUMA cable to connect the two units.

On the plus side, one unit is dual 800MHz, the other quad 800MHz. You'd need that NUMA cable, the rest is optional.

That looks good, I like it!

xiri wrote: Ok,
just compared the size of PROM v5.04 and v10.54 and it is exactly the same.

That means nothing. The capacity of the flash memory chip is constant also.

Congrats on your promotion to tall rack systems

I think a deskside SC will work too with a bit of handywork -- the difference is mostly the metal piece to slide it into the card cage.

Not sure I have a spare to let go, and I'm on the wrong side of the planet anyway. If you get desperate I can go digging ...

Ryan Fox wrote: It would be great to maybe host a site providing a video / content rich experience demonstrating each of these systems. From hardware , to booting and running demos

We'll see, maybe once it's back up and running. I also want to build my own site for my collection so I don't want to scatter all my content all over the web either, because I will have nothing left.

Back on topic: the system as I got it had an 8MHz IP4, but I still had a spare 12MHz IP4. Thought I'd give that a spin. Now, I've wondered whether the realtime clock of these things is powered by a small nuke, because the original clock still works in all of my Power Series, unlike just about every other SGI I own. But in this particular IP4 it was dead.

This is what I'm talking about:


Underneath that Sony CXK5816PN-15L static RAM (which holds PROM variables), a Dallas is hiding. It's a DS1216B (this photo is leeched from the 'net):


Basically this clock sits in a RAM socket and turns the RAM chip into NVRAM. Now, when it's dead, you have to key in all PROM variables every time the system powers on which is a major pain in the behind. You can still buy these things from Farnell while supplies last (they have 13 as we speak), but at nearly €50 a piece Should I buy one for my 4D/70, 4D/440, 4D/380 and Crimson, and maybe a spare for good measure? Nah, didn't think so either. Time to bring in the Dremel

I'm not a stranger to this job, but I like to know where to start grinding away. Sometimes the datasheets give you a clue where the Vbat wires must be hiding, sometimes you can find a photo on the internet. Not for a DS126, though.

Luckily, the wife of my colleague works in the Xray department of the local hospital

Which finally answers my questions about the longevity of these things ... not a nuke, but two batteries.

PS: since I had the opportunity, I had the Dallas chips of an Indy, Indigo2, O2/Octane and O200/O2K X-rayed as well. That will be the subject of another post.

miod wrote: The Bridge chip (and its offsprings XBridge and PIC) are PCI 2.1 compliant. It's only their IOC3 chip which is a fraud. If you don't deal with IOC3 in your PCI board experiments, you have nothing to worry about.

The PIC ASIC is is also the only bridge I've encountered which enforces the PCI-X spec to an annoying and pedantic degree.

I had to troubleshoot a PCI-X device driver in an Onyx3900 once, which mysteriously failed to translate DMA buffer addresses to PCI bus addresses. The IRIX function for that is pciio_dmatrans_addr() and you can tell it to generate 64bit PCI bus addresses with the PCIIO_DMA_A64 flag. Which we didn't because this was a 32bit device.

Except this flag was ignored on the Onyx3900 IX-brick.

Finally an engineer at SGI pointed me to this fine print in the PCI-X Addendum to the PCI Local Bus Specification, sect. 1.8, "Bus Width":

... and therefore PCIIO_DMA_A64 was always enforced on the PIC chip, whether you wanted it or not.

In other words: please fix your non-compliant PCI-X hardware. Which we did...

I almost universally use Seagate 15K.3 and newer series. They have FD bearings which are durable and quiet. They run much cooler than the disks originally supplied with most SGI systems. I usually use SCA versions with an adapter for 50pin systems.

They are not expensive.

zagnut wrote: Nope. More than likely, it will be a redundant RAID array for serving photographs.

IRIX does not do 'software' RAID, other than JBOD, RAID0 (striping) and RAID1 (plexing). The latter requires a license.

As far as hardware RAID goes: there are no hardware RAID controllers for SGI/MIPS systems. You'll need a disk array with built-in RAID capabilities, or a SAN server. Beware that there's a 2TB logical LUN limit in IRIX (this is a 32bit limit), arrays returning 64bit LUN numbers here will result in kernel panics and massive corruption. XFS does not have a 2TB limit, and XVM can be used to create large volumes out of 2TB or smaller LUNs.

I've got an old IBM DS4300 SAN server I occasionally use with my IRIX systems, it has ~ 8TB raw capacity (28*300GB). I can configure the disks in 2TB RAID5 LUNs and serve them up to an IRIX system which can then stripe them into a single, large volume using XVM (effectively a RAID50), and put a single XFS filesystem on it spanning the entire volume. SGI used to sell the same unit as the InfiniteStorage TP9300 and this is the way you were supposed to add large, redundant storage capacity to an SGI server.

The DS4300 is fully licensed and has many interesting enterprise features, but the noise level and power consumption of ~ 500W make this a silly thing to operate 24/7 at home. My current storage server (Supermicro, Haswell Xeon, 16TB raw capacity) idles at ~ 20W and is all but silent. And faster. Much faster.

noth wrote: Linux should be able to mount those XFS partitions and then you can aim john the ripper at the passwd file in last recourse. OS X can't mount XFS partitions.

Not necessarily.

Linux only understands the XFS v2 'on disk' format, the default since ~ IRIX 6.5.14. If these systems shipped from the factory with something older they will have the older on-disk format and Linux will not mount it, even if the Octanes currently run a more recent version of IRIX.

From the Admin> menu in the photo, choose '6' (sh)

You now have a shell in the installer environment

You can chroot into the harddisk using:

(this assumes the hard disk is mounted under /root)

Once you're chrooted into the system disk, issue

and change the password to your choice.

Type cntrl-D (or 'exit') twice and you're back in the 'Admin>' menu

Reboot the system and boot from hard disk.

Log in & enjoy

PS: Don't forget to create (and use!) a normal account for daily use.

kokoboi wrote: I have these files from AIX 4.1.4r4 for Motorola:

Which reminds me, there's one of these systems for sale, in NL:
http://www.marktplaats.nl/a/computers-e ... erver.html

Same guy has other esoteric UNIX hardware for sale, like an Intergraph clipper

If you're not interested in ATM networking: remove the board.

Otherwise, this sounds like an interrupt issue to me. VME boards to be installed left-to-right in the cardcage and you have to set or remove jumpers on the backplane of the system to configure the interrupt(s) used by the VME card.

Ian moved his site: http://www.sgidepot.co.uk/i2psu.html

Sounds like 2 hours of work plus a couple of bucks for parts -- that's how life is.

If it's anything like a car garage, they will supply the failed parts when they send back the PSU. Those parts could then be added to the shortlist of parts to replace by someone of the "I don't know what I'm doing but I know which end of a soldering iron gets hot" persuasion, in addition to "all small elcos".

You need at least a 286 CPU to run OS/2 v1.x. You should have bought an AT, not an XT.

Adobe software on IRIX uses a truly ancient version of FLEXlm (version 2, for Photoshop at least). The serial# itself isn't even used to nodelock the license, it's what was needed to convince Adobe to generate a nodelocked license for you and nothing else.

I sincerely doubt that a Creative Suite support site will magically spit out this ancient license format.

hamei wrote: Oh. I also know a guy here who was one of the "buy Chinese stocks, it's a booming economy, I tell you ! Just BOOMING !

Well, it went BOOM, right?

ivelegacy wrote: My definitive solution is: to rent a garage (or to buy it for cheap), then moving all my office over there.

That means you'll be away for the evening / day when you're messing with your computers which can quickly escalate to "I never see you anymore / your computers are more important than me / ...".

Also, in the end computers don't like to live in garages. If the humidity doesn't kill them, the unavoidable rodents will.

jpstewart wrote: Interestingly, the Linux version 5 info doesn't appear in the corresponding header (/usr/include/sys/fs/xfs_clnt.h) on IRIX. I wonder if IRIX supports it or if it's a Linux extension?

CRC support was added in Linux XFS, kernel 3.10 and newer. So it's relatively new and there's no IRIX implementation. At least in the current implementation, only meta data is protected with a CRC32 checksum.

sornywrx wrote: I just couldn't wait to join the forum and also needed to research and see if my SCSI HD will work with the Fuel.

I think that disk will work. But does the system come with the SCSI cable? The cable used in the Fuel has an unusual connector on the mainboard side.

I saw a "short" O3x00 rack half a year ago when I visited someone's place. It's stylish but not very practical as a general purpose 19" rack. It's vertical stands don't have the square holes for standard cage nuts for example, so you can't mount standard 19" rail kits in it. There was something else I think, IIRC it was less deep than a standard server rack. I'm pretty sure it's welded. I think almost everything except budget racks is welded.

I use a common 42U HP 10000 series G2 rack for servers and switches.

Pontus wrote: I might be so lucky to get a Crimson.

Congratulations to you, sir!

Pontus wrote: Should I even try, the locking mechanism looks potentially brittle. Will removing the skins risk doing more damage than I might cause by transporting them left on?

If you know what you're doing it's pretty straightforward, but it depends. I used a trailer to transport my Crimson and only took the front door off. I did wrap it in thick moving blankets and strapped it down, though.

Pontus wrote: is it brittle like the Indigo2 or Onyx2?

No, fortunately it's not because removing the skins is a rather "physical" act. It goes like this:

1. The front door (the entire assembly, not just the drive bay door) can be removed by prying a big flat screwdrive blade in the spring-like upper hinge. This forces it up and the door panel "falls out".

2. The skirts can be gently pried off.

3. The large side and back skin panels are attached with half a dozen or so of the black clips you see in Pentium's photo. I tried to gently undo them one by one but you'd have to flex the side panel in impossible ways. Instead, I usually crouch by the side I want to remove, place my foot against the chassis where the skirt was (us the metal handle for the back side), place my hands under the panel and with a swift pull you pull the entire panel off, towards you. This can be intimidating because it probably requires a bit more force than you feel comfortable with and that's why I'm happy they are *not* brittle. The top of the side and back panels goes underneath the top cover but it's not stuck there, it falls right out.

4. Finally the top. On the right hand side (looking from the front) there are two clips in the metal chassis, underneath the top skin. I forgot if they have to go (a bit) up or down, but they keep the top in place. The top comes off towards the front (or was it the left?) and again this can take a bit of knocking with an elbow or rubber hammer.

The black clips came in two flavors, the one you see in Pentium's photo is usually what you find in Crimsons and Onyxes. Older single tower PowerSeries had more rugged clips. The four 'fingers' seen in Pentium's clip can easily be bent out of shape if the panel is not 100% correctly aligned when you re-install the skin panel. This is mostly a problem with the side panels. The back side skin has alignment 'studs', and the skirts simply take a lot less brute force to put back. Fortunately the skins will stay on with half of the clips semi-broken or missing, but if you really run out I have some spares. Seems like something you could easily make yourself with a 3D printer as well.

Other hints for moving PowerSeries and Crimsons: they have wheels, but only roll forward/backwards. So if you place it sideways in a car (station wagon) and fill up the space with boxes etc, it won't go anywhere unless you have an accident. If you don't place it sideways it could knock the back window out of your car. The handle on the back of the chassis is strong enough to carry the weight of the system (on that side). Oh, and you might want to remove the disks before moving the system.

ivelegacy wrote: I prefer Solid Impact : it eats less electricity, and it makes the machine less warmed

I used to use an R10K Indigo2 as an IRIX 6.2 build system. It was optimized for that task: no graphics at all, but a G160 fast ethernet board. Big, near-silent disk and 1GB RAM.

I guess that effectively makes it a PowerChallenge M or something.

Since this is about an Indigo it's worth mentioning that EFS, the file system used by IRIX 5.3 and earlier has a maximum size of 8GB, and R3000 Indigos do not run IRIX 6.x.

Yes, you can create multiple slices of 8GB and mount them as /usr, /var, usr/people etc. etc. but this quickly becomes rather silly. And IRIX 5.3 w./ XFS exists but it has it's own quirks, especially on pre-R4000 CPUs.

All these converters

I've got an APC 7953 metered PDU behind my desksides / workstations, and another in my 19" rack. Remote power switching made easy
I'm using a Cyclades ACS48 terminal server.

All of this is designed to do this job and it does it well. It can even be scripted but I can't be bothered. The biggest downside is the amount of serial cable you end up when you implement this for two dozen systems, when you rarely need serial access under normal conditions.

For sale: last rev. FP1 with Heart rev F (needed for bigger VPro graphics I believe).

'hinv' extract from actual part:

You can have either the whole assembly including the metal front, the SCSI backplane and fan, or just the FP1 board to save shipping.

It passes all diagnostics.

Price 75EUR + shipping (from The Netherlands)

I have an Octane2 CPU board I can do without.

Passes the IRIX diagnostics and comes with a 300MHz R12K CPU and 512MB RAM.

Price: 75EUR + shipping from The Netherlands

Qty 3 available.

I think they're fully functional, but they're probably most useful as a latching mechanism donor to convert an Origin/Onyx2 XIO card to work in an Octane. They have good compression connectors as well.

10EUR each, plus shipping from The Netherlands

This IO6 is still sealed in an ESD bag and appears to be brand new.

I don't have an Origin 2K to test it anymore.

10EUR + shipping from The Netherlands

Part# 030-1286-002

Tested with IRIX diagnostics, fully functional.

EUR25 + shipping from The Netherlands

Type "green with new style logo"

Just the green plastics parts. I'll take pictures if you want them, but I promise you there's not a scruff or scratch on them. Of course the door mechanism works too. Both drive bay blanks installed.

This is your chance to fix up that poor Octane that was abused by the recyclers / postal service / ...

50EUR + shipping from The Netherlands.

Reserved

Trippynet wrote: Every single Indy I've ever come across has not had a Floptical drive, but has had the hole left open. It's just part of the design.

Wrong.

Normally you'd have a cover plate which can be "installed' in the top drive bracket, and when the drive bracket is installed in the system it closes the floptical drive hole.

If you look carefully at this photo you can see it somewhat (well, maybe only if you know what to look for) It's just a piece of sheet metal, like the drive bracket.
http://www.sgistuff.net/hardware/system ... erior1.jpg

mapesdhs wrote: Don't bother using the PROM ide for testing, it's not accurate, often reporting false positives.

Since we're doing thread necromancy anyway: I'm afraid I have to disagree with this statement.

<engineering hat on>

I repaired a pair of TRAM modules last year . They failed the IDE diagnostics. I put them under the microscope and reworked failed soldering joints until they passed the diagnostics. There were a lot of failed joints. Regardless, the demos worked just fine on these obviously broken TRAMs. Conversely, I've never seen a system pass the IDE but visually fail a demo. So: I'll trust the IDE, and not the demos.

How do you prove a "false positive" of the diagnostics anyway? Is it a sentiment of "I don't see anything wrong"? Because that logic is flawed unless you're 100% sure that you have exercised every functional unit of the hardware. Which is basically what the diagnostics are for in the first place. The texcube demo runs the same (on MaxImpact) with or without the TRAM option: it simply requires only a small amount of texture memory. So while artifacts on the screen are proof of failure, lack of artifacts doesn't prove anything about the functional state of the hardware.

<engineering hat off>

The biggest problem I have with the diagnostics is that they're strictly field diagnostics: the output is a boolean (pass/fail). I would have liked more verbose output along the lines of "TRAM chip at PCB location C2D4 failed test X at address Y". I can put an IMPACT TRAM under a microscope and inspect a failed board chip by chip, one pin at a time, but for something like an RM4 this is simply not feasible. Surely they had something like that in engineering but alas....

Krokodil wrote: Is it actually possible to fix the things? You could make some good money if you did.

Yes, I had some success least year .

The whole procedure is not for the faint of heart. It is also very labor intensive so it would almost certainly cost more than most hobbyists are willing to pay.