"Workstation"-class machines have long been available in today's conventional desktop form-factor, but have also been available in a rack-mountable form-factor, for installation into a server-rack in another room in the building - or in a "deskside" cabinet.

Going right back to the early 1980s, when CAD systems would run on systems like the Data General Eclipse, which would necessarily have to be in another room, this meant there would have to be some connection from the user's desktop to the computer next-door - but also through the 1990s and early-2000s with the SGI Onyx and Tezro respectively.

(Tangentially, I'm wondering why they're even called "Workstations" if they live in a rack: usually it's only headless servers that can really be accommodated in a rack; while SGI did have other rack-mountable machines specifically called "Servers" and not Workstations, so is there anything to the naming convention? In my head the term "Workstation" implies it's being used for real-time interactive use by a single user, whereas some of the documentation for the rack-mount Onyx implies it might be used in a TV production situation, generating graphics in real-time directly for broadcast driven by automated software instead of someone interactively using CAD/CFD/3D software).

So assuming this rather beefy hardware is running in a rack on the other-side of a wall, how are the end-user's mice, monitor, and keyboard, and any other peripherals (e.g. CAD digitizer), supposed to be attached?

...in the case of D-Sub VGA, I understand the maximum cable length is on the order of 150 feet, but given these computers' displays would be running very high resolutions, sometimes even by today's standards, certainly not capable of being transported by an early-1990s VGA cable, so how would they have been connected over, say, a 75-foot distance? In SGI's case they had their DB13W3 connector and cables but I can't find any information about its maximum-length, and Wikipedia says that D-Sub VGA and DVI effectively replaced it anyway, but DVI has a very short maximum cable length (~5m, I understand) - so if DB13W3 did support cables long enough to stretch between rooms it seems odd that it would be replaced with an interconnect that couldn't.

Mice and keyboards also pose a challenge: even if USB existed back then the USB 1.x specification sets a maximum length of 3m - you can workaround this using powered repeaters/hubs but that still wouldn't get you beyond ~15m, and bulky hubs wouldn't work with tight conduits between rooms - not to mention hub propagation latency making mice awkward to use. As for PS/2, I couldn't find any authoritative info on maximum cable-length.

So imagine it's 1992 and you work at Pixar and you've got a flashy new rack-mount SGI in the HVAC'd room next-door - how does your mouse and keyboard, and (multiple?) monitors connect to the box?

And if you've got a CD full of art assets you want to load-up, do you have a desk-top SCSI drive with a cable snaking its way next-door - or would you need to physically walk to the rack-mount box and pop the CD in that way? If there's dozens - or more - of people with similar workstations how do the IT people manage having to route all those (presumably very thick) cables?

  • 11
    X terminals and thin clients were common up into the 90s. It was effectively a small computer with Ethernet, display, keyboard, mouse, and enough CPU/GPU/memory to run a TCP/IP X11 server, but no disk and no real operating system (often netbooting as Brian mentions). If you had one on your desk, then you could use it for login sessions on any Unix machine(s) on the network. Could be a big shared server, or a rack-mount workstation that you got to use exclusively. Commented Mar 12, 2023 at 5:57
  • 5
    Where did you hear that workstations were rack-mounted? Sure, there (eventually) 1U servers, but workstations were always on or beside the desktop for the very reasons you mentioned.
    – RonJohn
    Commented Mar 12, 2023 at 10:02
  • 5
    be careful of the term "workstation," as the definition is very nebulous. What a "workstation" is and what is does varies wildly, even by the same manufacturer, in the same era. Heck, even the definition of the term "computer" will vary back then.
    – Keltari
    Commented Mar 12, 2023 at 19:18
  • 2
    Sounds like the existing answers are about Serial and muliti-user timesharing hosts, whereas @dai's question is specifically about a high-powered single-user graphical workstation that is physically in another room. Does that narrow down your interest range ?
    – Criggie
    Commented Mar 13, 2023 at 3:40
  • 2
    @Criggie Yes, that’s my question - I certainly waffled on in my post.
    – Dai
    Commented Mar 13, 2023 at 16:06

8 Answers 8


Another route was X terminals: low-powered computers with big screens that communicated with a CAD server over Ethernet. Keyboards, mice, digitizers and other peripherals were connected to the X terminal, by whatever cabling was used at the time - often RS-232. The X protocol carried information from peripherals to the CAD server over Ethernet; it's quite general-purpose.

  • 19
    @Dai You can't? X11 worked pretty decently with 1600x1200 on 10Mb Ethernet back in the days. With 100Mb it was flawless.
    – vidarlo
    Commented Mar 12, 2023 at 15:19
  • 12
    @Dai it worked fine. Actually, I should say "it works fine". X does have its problems, but network latency and bandwidth has never been one of them and it's still not totally obsolete.
    – JeremyP
    Commented Mar 12, 2023 at 17:07
  • 12
    @Dai: It worked perfectly well, and still does, if you're not using some GTK3 monstrosity that's constantly streaming raster graphics for UI elements over the wire. X was designed to do this, and does it exceptionally well. Commented Mar 12, 2023 at 18:11
  • 10
    @Coxy I'm not sure you've experienced CAD applications in the late 90s. That "unplayably low framerate" was what we're used to when using CAD. Instead of FPS we measured performance in seconds per refresh. As in, 5 seconds per refresh (of entire drawing) was high performance for a complex Autocad or Protel drawing and 2 second per refresh was "workstation class". 10MB Ethernet have enough bandwidth to transfer line drawings (basically similar to modern SVG files) within 2 seconds.
    – slebetman
    Commented Mar 13, 2023 at 11:15
  • 5
    @marcelm All questions of how well remote X11 works today aside, the fact of the matter is that it did work acceptably well in the 1990s, even for graphic-intensive applications, over 10Mbit Ethernet and distance scales on the order of a college campus.
    – zwol
    Commented Mar 13, 2023 at 19:22

In 1992 my workstation wasn't rackmounted, it was a pizza box under my CRT monitor or a PC-style tower or half-tower tucked under my desk. Servers in the machine room were the same form factors (pizza boxes, towers, lunchboxes) but headless. The machine room also had a rolling cart with a 17-inch CRT, a keyboard, and a mouse, but it was fairly rare to plug them into anything. A couple years later KVM (keyboard/video/mouse) switches started to appear, but my workplaces tended to not have them. There were often Wyse 50 or 70 terminals in machine rooms because they were cheaper than DECs and could connect to the serial consoles of the headless servers.

  • 1
    Certainly by the late 90s (and probably earlier), there were KVM extenders that enabled control of a racked computer from another room, hundreds of feet away. Commented Mar 13, 2023 at 2:21
  • 1
    I remember the day when we switched off some old NT4 workstations and a Solaris pizza box in a 8-person room. It was very weird to hear ... nothing ... after years of daily tornado sounds!
    – AnoE
    Commented Mar 14, 2023 at 12:29

on systems like the Data General Eclipse, which would necessarily have to be in another room

Not necessarily. In the link, they talk about "Series 1000" and "Series 2000", so given the year, I assume they mean an Eclipse MV/1000 and MV/2000, which are part of the MV/8000 line.

We have pictures of how that looked like, e.g. in the picture from here, which shows an MV/8000 installation (from Data General sales literature, note the absence of cabling, but you get a sense for the scale of the equipment):

An MV/8000 installation (from DG literature)

You can see the cabinets against the wall and the "workstations" (screen with keyboard) in the middle of the room, with people sitting in front.

why they're even called "Workstations" if they live in a rack

Well, the minicomputer lives in a cabinet, the "stations" where people sit down and "work" don't. It even says in your link that e.g. the MV/1000 "supports two workstations".

end-user's mice, monitor, and keyboard, and any other peripherals (e.g. CAD digitizer), supposed to be attached?

Just like on your PC: Through a cable that goes to specific peripheral hardware inside the cabinet.

For example, there is the Data General Dasher line of display terminals for the earlier Nova, connected via a serial link with up to 19,200 bps. The D460 had graphics capabilities via up to 3572 user-defined characters or symbols; this was e.g. used in the Data General Trendview and Present software packages to display pie charts etc.

I have not found a catalog of the MV/8000 periphery, but they probably had similar graphics capabilities inside the terminal, which is enough for CAD.

in the case of D-Sub VGA

The graphics are generated in the electronics integrated into (or close to) the monitor, not in the cabinet.

Mice and keyboards also pose a challenge: even if USB existed back then the USB 1.x specification sets a maximum length of 3m

Serial can easily go looong distances. Just look at how IBM terminals are connected to the mainframes. And no, it wasn't USB. :-)

  • 3
    IBM terminals were connected to Front End Processors, and they were connected to the mainframe. Very similar to (but faster and more efficient than) the terminal servers which dumb async terminals plugged into.
    – RonJohn
    Commented Mar 12, 2023 at 10:05
  • 5
    Just keeping in mind that the image appears to be a marketing staged image. There isn’t an inkling of a cable anywhere between the terminals and anything, not even power. But I accept that the general layout would have been typical (presumably with actual desks instead of lightweight conference furniture) Commented Mar 12, 2023 at 15:41
  • 2
    @EuroMicelli I've also wondered about the cables :-) But it gives a good idea of the size of the equipment, and why it's perfectly fine to have them in the same room (unless you want to keep the noise out).
    – dirkt
    Commented Mar 12, 2023 at 17:55
  • 2
    @EuroMicelli - Heh, I didn't notice the lack of cabling, but one look at the perfect posture of the seated guy in the white suite screamed "that's not real" to me. :-)
    – JonathanZ
    Commented Mar 12, 2023 at 18:45
  • 1
    @SimonCrase That's the "foot" of the legs of the unoccupied chair in front of the desk she's at.
    – Dai
    Commented Mar 14, 2023 at 0:54

I have seen a setup where the machine room and the terminal room were neighbors and connected through a series of holes in the wall. Machines were in racks in the machine room and keyboards, screens and mice were in the terminal room. Max cable length was probably less than three meters.

The point of this setup was to avoid noise, and also cool the machine room down below human-comfortable temperature.

I doubt this was a common setup.


There are two answers.

Answer 1. This was rare: although machines with framebuffers sometimes came in rack-mount form they were generally then being used headless. Far more common was something like a Sun 3/160 which was a fairly substantial VME-based machine which sat in a deskside box. There was an entirely (from memory) equivalent machine, the 3/180, which sat in a rack. We had both: the 3/180 was in the machine room with a terminal as its console and the framebuffer unused, and the 3/160s in the office with screens. And yes: they were not that quiet, but quieter than the 3/180 with a couple of Eagles underneath it. People put up with a lot of noise.

Answer 2. When it was done the answer was long, often custom-made cables combined with often marginal display quality. As an example we had a Symbolics 3670 which was a full-height 19-inch rack (not, I think, completely full, but the machine itself was perhaps half of the rack), and the disk (another Eagle) was another good chunk of it). Nobody wanted that in the office, even then. So it sat in the machine room with a very long cable running up 4 floors (I think) to its screen, keyboard & mouse. I think the mouse & keyboard plugged into the back of the screen: obviously the entire interface was proprietary. The screen was marginal but usable, obviously helped by being B/W.

  • 1
    It was possible to rack-mount workstations in these enclosures, though. We ran a number of Sun 3/60s in a shared 3/160 enclosure to service modem lines. (Both the 3/50 and the 3/60 used the third VME Bus connector on the 9U Backplane for power supply only, so they could share the enclosure and mighty fan array without problems) Commented Mar 15, 2023 at 13:02
  • @TatjanaHeuser: I can't remember seeing that done, but it makes sense, since those machines were just a single VME board inside, weren't they. Do you know if Sun supported that, or did you just do it as a neat hack (which it clearly was!)? Commented Mar 16, 2023 at 12:22

So imagine it's 1992 and you work at Pixar and you've got a flashy new rack-mount SGI in the HVAC'd room next-door - how does your mouse and keyboard, and (multiple?) monitors connect to the box?

I never worked at Pixar but can tell my experience with SGIs. My college had SGIs (Indys?) as workstations in a lab at the third floor which were connected through ethernet (10Base; the department did not have 100Base-T network when I was there) to the larger SGI server (could have been an octane) and the maspar (this one was also in the third floor but in its own room; I can't remember where the SGI was server). You logged into the workstations, which would NFS mount your homedir, and then if you wanted to run something in the servers you started a Xsession in it (think VNC) or xforwarded the application into your workstation.

That was what made X11 better than any GUI at the time (Mac, Windows, Amiga). Playing a multiuser combat simulation was very easy compared to the PCs, and SGI did come with one.

FYI, those machines were bought with grant money, so years after grant was over they were used for some undergraduate classes. Ok, not the maspar; that was for grad research. The Indys were not as loud as the desktop PCs the faculty had.

So, what about the rodents and keyboards? Every machine had their own. The server monitors were usually small crappy ones because they were only turned on if the sysadmin needed to do something to them in person. The Maspar keyboard usually sat on the top of the rack.

  • "Playing a multiuser combat simulation was very easy compared to the PCs, and SGI did come with one." - realtime fully-3D rendered, with texturemapping too? How would that have been playable over X11 on a 10mbps network... unless your window sizes were tiny (like 320x240-tiny?)
    – Dai
    Commented Mar 14, 2023 at 0:41
  • 1
    @Dai X11 is more high-level than you might think; it doesn't work like VNC or RDP. In particular, OpenGL can be tunneled through X11 and rendered server-side (i.e. user-terminal-side).
    – Sneftel
    Commented Mar 15, 2023 at 17:51
  • @Sneftel Right, but doesn't terminal-side OpenGL rendering defeat the purpose of having an expensive rackmounted graphics workstation? Also, by the mid-1990s a 3dfx Voodoo running Quake is going to be piping a lot more than 10mbps (Fast Ethernet) over the PCI bus - then add-in network latency - it just doesn't add-up to me...
    – Dai
    Commented Mar 16, 2023 at 21:09

One company I worked at used video/keyboard extenders that allowed the machine to be upto about 50 metres from the monitor. It did cut down on fan noice in the office but required permission from the IT manager to reboot your machine and 2nd monitors where not possible.

If I knew the setup before took the job I would have not taken the job as companies with such control attudues where not fun to work for.


There used to be several options, depending on if you really wanted to access the framebuffer of the workstation for its graphics:

  • Console-only, wired to a serial terminal via a (sufficiently long) serial cable.
  • Console-only, serial console port wired to a terminal concentrator (mapping its serial ports to telnet port adresses).
  • Framebuffer ("Graphic Card") wired via an extension "cable" (some of those translating to fibre optics as a means to extend the limitations of the default cabling). Not cheap. Keyboard and mice used to be serial, so these were usually less of a limitation. (Of the variants listed here, this or a full-fledged KVM extender was the only one providing access to the graphic console of the system)
  • For more cumbersome systems that did not offer a serial console port, KVM extenders had to be used, and still are in use today.
  • XDMCP login to the system from any other desktop system running X11.

When Workstations were mounted in racks, that was often done to ensure better ventilation for the system and less noise for the user. People would rarely connect directly to a framebuffer, even if present. For those that needed the framebuffer, long cables were available (at a price). The keyboards were usually serial, so using a longer cable wasn't a problem. More often the (serial) console would be used though, either with a terminal attached directly, or via a terminal concentrator that bundled 8 or 16 terminal lines, supplying them via telnet ports to everyone with access to the subnet in question.

Those desiring fancy graphics most often used X Terminals or older systems running a stripped down OS presenting an XDMCP login.

  • 1
    The question asks "how", but your answer starts with "that". What do you mean by "that" in the first sentence? A specific technique? (You can't mean the concept of wiring things up.)
    – wizzwizz4
    Commented Mar 15, 2023 at 16:54
  • Re: "serial keyboards" - can you name some keyboard connection/connector/interconnect standards or specs that would have been used around then? Would they have been AT or PS/2 - or some other spec? I understand AT and PS/2 are both technically "serial" but they don't use a serial spec that would allow for long cable lengths).
    – Dai
    Commented Mar 16, 2023 at 21:17
  • I'm still holding onto my Sun3 keyboards - serial DB9 was very much en vogue back then, and they're still doing great. DEC used some kind of twisted pair ("telephone") connector, but with an asymmetric offset to the locking mechanism. Commented Mar 19, 2023 at 16:16

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .