[TUHS] Whither Workstations? (Was Re: Demise of AT&T)

[Kevin Bowling]

On Mon, May 19, 2025 at 1:28 PM segaloco via TUHS <tuhs at tuhs.org> wrote:
>
> On Monday, May 19th, 2025 at 12:45 PM, Dan Cross <crossd at gmail.com> wrote:
>
> > On Mon, May 19, 2025 at 12:36 PM Noel Chiappa jnc at mercury.lcs.mit.edu wrote:
> >
> > > > [snip]
> > > > It wasn't just AT&T, IBM & DEC that got run over by commodity DRAM &
> > > > CPU's, it was the entire Minicomputer Industry, effectively extinct by
> > > > 1995.
> > >
> > > Same thing for the work-station industry (with Sun being merely the most
> > > notable example). I have a tiny bit of second-hand personal knowldge in this
> > > area; my wife works for NASA, as a structural engineer, and they run a lot of
> > > large computerized mathematical models. In the 70's, they were using CDC
> > > 7600's; they moved along through various things as technology changed (IIRC,
> > > at one point they had SGI machines). These days, they seem to mostly be using
> > > high-end personal computers for this.
> > >
> > > Some specialized uses (various forms of CAD) I guess still use things that
> > > look like work-stations, but I expect they are stock personal computers
> > > with special I/O (very large displays, etc).
> > >
> > > So I guess now there are just supercomputers (themselves mostly built out of
> > > large numbers of commodity CPUs), and laptops. Well, there is also cloud
> > > computing, which is huge, but that also just uses lots of commodity CPUs.
> >
> >
> > The CPU ISAs may be largely shared, but computing has bifurcated
> > into two distinct camps: those machines intended for use in
> > datacenters, and those intended for consumer use by end-users.
> >
> > CPUs intended for datacenters tend to be characterized by large
> > caches, lower average clock speeds, wider IO and memory bandwidth.
> > Those intended for consumer use tend to have high clock speeds, a bit
> > less cache, and support for comparatively fewer IO devices/less
> > memory. On the end-user side, you've got a further split between
> > laptops/desktop machines and devices like phones, tablets, and so on.
> >
> > In both cases, the dominant IO buses used are PCIe and its variants
> > (e.g., on the data center side you've got CXL), NVMe for storage is
> > common in lots of places, everything supports Ethernet more or less
> > (even WiFi uses the ethernet frame format), and so on. USB seems
> > ubiquitous for peripherals on the end-user side.
> >
> > In short, these machines may be called "personal computers" and they
> > may be PCs in the sense of being used primarily by one user, but
> > contemporary data center machines have more in common with mainframes
> > and high-end servers than the original PCs, and consumer machines are
> > much closer architecturally to high end workstations than to
> > yesteryear's PCs.
> >
> > My desktop machine is a Mac Studio with an ARM CPU; I call it a
> > workstation and I think that's pretty accurate. At work, one of our
> > EE's has a big x86 thing with some stupidly powerful graphics card
> > that he uses to do board layout. It's a workstation in every
> > recognizable sense, though it does happen to run Windows.
> >
> > - Dan C.
>
> This may be getting into the weeds a bit but don't forget industrial hardware, the stuff where you approach the blurry demarcation between CPU and MCU.  This for me is always the third class when I'm discussing that sort of thing.  Of course this also means "operating systems" closer to standalone applications sitting on top of some microkernel like (se)L4, but you did have VME-based workstations and UNIX versions specifically for VME systems.  For me these walk a line between true workstations and minicomputers, but as usual that is from the perspective of having not been there.  For the record, one of the canonical UNIX development environments from AT&T for WE32x00 stuff was a WE32100 and support chips thrown on a VME module running System V/VME.  From what I know, VME is still quite common in industrial control.  How much UNIX and workalikes constitute the OS landscape in today's VME ecosystem eludes me.

I would not be surprised if the VME board was used to prototype
datakit stuff.  A major benefit is the eurocard form factor, which is
easy to build custom hardware around due to a shared ecosystem.
Physics labs commonly spin their own boards, you will find a lot of
VME docs online hosted by them.  The VME design is relatively "clean"
and "simple" while still allowing for quite a bit of power, and it was
pumped over the years to 320MB/s while a center P0 connector could
hold any desired newer busses for unbounded interconnection.  One area
it shines is shared memory multi-computers, nothing as popular
thereafter really encourages this way of systems design.

But the major "success" of the WE32k was the 3B2.  Those sold in good
number.  I don't think the WE VME product did from my own research.

As others have alluded to, VME was once popular in UNIX workstations.
Sun3 and early SGI are the popular examples but there were others like
Motorola and many smaller and niche players.  Both Sun and SGI had
ways of bringing VME along once newer busses took over.  For instance,
SGI is common in flight simulators and a VME crate would drive the
real time interfacing and control while the SGI orchestrated the
situation and graphics.

> Either way, I feel like this is a class of computers that frequently flies under the radar, but if we're talking strictly consumer stuff, yeah VME very quickly loses relevance.

VME is pervasive enough there are still "new" designs, although they
are dwindling as switched and high speed serial protocols have taken
over all aspects of system interconnect.  VPX is the new one.  But the
tail is long, any new or overhauled M1A2 Abrams tank is getting some
conduction cooled VME hardware for instance.

> - Matt G.