Let's see if you're interested in some more computer-pr0n

Warning for the bandwidth starved: these picture total nearly 5MB.
This is my 4D/380 VGX system. It's a "Predator rack", and it used to belong to one of the German Fraunhofer institues where it was used for climate modeling.
It has the awesome compute power of eight Indigo R3000 systems and a ferocious appetite for electricity
The pictures are old, I took them in 2008 when the power supply of the system broke. When I fixed this, I decided to tear it apart completely and clean it thoroughly before putting it back together. This was before the construction of my computer annex so this system was in my garage.
A tour of the outside
The system is roughly 1.60 ~ 1.70m tall:
Some desksides for scale:
There are two doors on the front side. The top one covers the card cage access panel with all connectors, the bottom one the drive bays. Notice how only the top one-third of the system is the actual computer. The grille underneath the card cage is where the blowers suck in fresh air (air exhaust is out the top of the rack). The cage in the middle has the control panel with the Power Meter blinkenlights. In the bottom of the rack are two more 19" shelves for extra disks.
A closer look at the connectors on the card cage access panel. Graphics output is with BNCs; I've installed a 13W3 adapter. The two black cables connect the IPI-2 controller with the Seagate ST81154K in the bottom of the rack. To the left some connectors for digital and analogue video I/O using a separate VideoLab breakout box. The silver DB15 is the keyboard. On the far left you can see the two MIC connectors for FDDI (100Mb/s network using fiber optics). The DB9's are all serial ports.
A better look at the control panel with the system disk (5.25" full height SCSI). Mine has a CD-ROM drive, most would have had a QIC150 cartridge tape drive. When I got the system the tray with the SCSI devices was missing; I obtained one later. The mode switches can be used to debug hardware problems much like the MSC switches on an Onyx2.
Mandatory shot of the infamous "Power Meter" blinkenlights
A better look at the drive bays in the bottom of the rack. Mine has a 1GB IPI-2 disk (Seagate ST81154K) in the bottom-left quadrant. IPI-2 was a high-end competitor to early SCSI. This thing is a monster; it basically occupies half a 19" shelf, weighs ~ 30kg, has it's own 230VAC power supply and the touch panel can be used for maintenance independent of the system: spin or park the disk, key in bad sectors etc etc.
Back of the rack with the access door open. You can see the keyboard and video extension cables and the breakout for the VideoLab video option. In the bottom the PDU, and you can still spot the back side of the Seagate IPI-2 disk.
A view of the inside of the card cage. In the middle you can see four IP7 CPU cards, to the left of it the IO3 board and to the right of it the MC2 memory card. This is the "compute" part of the system. The right half, left to right: the GM and GE boards, the VO1 (VideoLab option, black handles), a pair of RMs and the DG card. On the far left some VME option cards: a Xylogics IPI-2 controller and a VME FDDIXPress card (with the orange fibers).
Just to show you how big this IPI-2 disk really is. Power supply at the front, the read 2/3rds is the actual disk. This is an 8" disk.
A view of the card cage blowers from the back of the system after the backplane was removed. This is why you can put a four-RM Reality Engine in this system and not in your deskside...
Next: the internals
The internals
The backplane. Main 5V bus bars in blue. The power supply is rated for 200A @ 5V. The bus bars were laminated by ... Durex

Aux power connectors in the bottom left, plus connectors to hook up the SCSI and control panel in the center bay of the system.
A look at the VME jumpers. Unlike the Professional IRIS before it, the PowerSeries are less critical when it comes to VME jumpers, but in some cases you need to remove them.
The internals
The IP7 CPU board. At the front two SM1 daughtercards, with an R3000 CPU, R3010 FPU, some cache memory and glue logic. This modular nature makes the IP7 CPU boards quite serviceable: it is possible to swap SM1 modules or CPU chips between them and rebuild one working IP7 from two broken ones. In the back the PowerPath MP bus logic, and a set of boot ROMs for every CPU. All IP7s in a system must have identical boot ROMs. When I got this system, it had been used as a donor for another system. It had two faulty CPUs, and ROM mismatches. The primary CPU was faulty which made the system really unhappy. It looks like the FPU on the right SM1 of this IP7 has been replaced once, because it lacks the copper lid.
The MC2 memory board. A PowerSeries can have a maximum of 256MB of RAM. The SIMMs on this board are proprietary and exist in a high- and low density version. An MC2 populated with low density SIMMs has 64MB, the high density SIMMs add up to 256MB. It's possible to mix low- and high density SIMMs on an MC2, but memory must be installed in banks of 4 identical SIMMs iirc. It is possible to install more than one MC2 in a PowerSeries rack, but mine has a single MC2 with 256MB of high density SIMMs installed. This amount of RAM alone would have bought you a nice house in 1990.
The IO3 board. It has two SCSI channels (one via the backplane), AUI 10Mb/s ethernet. In the bottom right quadrant you can see some audio hardware, but to the best of my knowledge it was never made available by the software and it isn't wired. There are four IO PROM chips, the version must match the CPU PROMs. Early versions of the PowerSeries PROM were built from the IRIX 3.x code base, but the last version was built from the IRIX 4.x code base and will allow you to boot from a CDROM.
The GM3 Geometry Manager is what connects the GE board to the PowerPath MP bus of the 4D system. It has a Motorola 68K chip. You can interface with it using the DB9 seen at the front of the board.
The GE6 board of the VGX set. You can clearly spot the eight GE chips (the ceramic tiles). It's connected at the front to the GM board.
One of the two RM boards in this VGX. It will work with one RM, this is called a 5-span configuration. The dual RM version is called a 10-span. Each RM has twenty Image Engines. The RMs and DG are connected by a bridge board at the front.
The VGX DG board. You can clearly spot the Bt RAMDACs. The open space in the bottom-left quadrant is for an optional daughtcard. For example, my 4D/440VGX has a BVO (Broadcast Video Option) here which converts the graphics output to composite SD video. This system has the much more powerful VideoLab hardware which is a separate board and breakout.
And this is the VO1 VideoLab board. Just look at the amount of custom hardware on this board. You have to realize that video hardware from this generation is usually "single frame" hardware: the system would render one frame of video output, the video option would output this one frame, and send a signal using VideoLAN to a capable video deck or abacus unit to record the frame. Rinse and repeat a couple of thousand times and you would have a couple of minutes of animated video. This would take all night of course.
Systems from this era didn't have the system or I/O bandwidth to deal with realtime (streaming) video. But the VGX has the graphics bandwith. The VideoLab can do tricks like the O2 can: the CPU animates some geometry (e.g. a cube) and the VideoLab put a live video stream a texture on top of it. It can output the result as video (like a GVO option in an Onyx2) and you can see it on the monitor. The VO1 has inputs and outputs for component-analogue video and "601" (parallel) digital video. I have converters from parallel digital to SDI, but haven't had a chance to piece it all together and give it a spin. This is definitely on my to-do list
The FDDIXpress VME option board. This would give you 100Mb/s networking instead of the standard 10Mb/s provided by the IO3. I have FDDI infrastructure (concentrator, router) so I'm actually using this.
And finally: the xylogics SV7890 IPI-2 controller. This connects to the big Seagate disk in the bottom of the rack. This controller is supported by the I/O PROM on the IO3 so the system can boot from it. I have IRIX 4.0 on the IPI-2 disk and IRIX 5.3 on the SCSI system disk, and can boot either by changing some environment variables.
That's all. Hope you liked it.