Large inventory of transputer parts for sale!!!

From: Gary Oliver <go_at_ao.com>
Date: Thu Jul 22 02:32:35 1999

At 02:10 AM 7/22/99 +0100, you wrote:
>snip


>I've never worked with T9000s, but they look nice chips from the
>databooks that I have. But although they are clearly related to the
>T2/T4/T8 series, there are enough differences that I could consider
>calling them 'non-classic'. It's up to you, though.
>
>-tony

Well, I never worked with T9K processors either, but I do distinctly
remember my awe-turned-revulsion several years ago when the reality
of the T9K became clear. (This is visceral from memory - if you
need details I'll have to get out my T9 books.)

Many years ago (in the heyday of the T4 and T8) I co-authored a C
compiler (Logical Systems) for the T2, T4 and T8 processors. I've
hacked a lot of code over the years - even written a couple of other
C (and weirder) compilers, and the Transputer was one of the most
elegant machines I ever wrote code for. (I also liked the National
320xx machines, but they never were fast enough to compete with
anything real - but that's another story.)

 From a compiler-writer's standpoint, the elegance of the T8 (and family)
was a dream. (Thought the variable-length instructions and it's impact
on optimization was a bit of a hassle, but the compiler got to at least
push the "jump" optimization off to the linker and that wasn't my job!)

I attended two (maybe the last two) Transputer User's Group conventions,
and in the first of these, everyone was abuzz with news of the T9s.
But the chips were delayed and delayed and by the time of the second
convention, many of the "ooohs" had turned into "ewww" (the sound one
makes involuntarily when driving past a dead skunk on the highway.)

Later when doing a little consulting to users of T8s, I did a small
job for a large wall-street company that wanted to move a process they
were running overnight on their "big expensive" IBM mainframe and place
it on a small farm of T8s. After working out a process for distributing
the computation (a BIG matrix cruncher) over about 16 T8s, a trip was
made to the customer's site. We merged our software and started a "test
case" before going to lunch. However, it finished almost immediately
(about 5 minutes.) We thought "oh crap.". However we looked over the
results and it had produced the correct results. The "test" was a run
that had previously taken several hours on their IBM mainframe.

This was on a bunch of 20mhz processors (if I recall) so the net performance
of this SIMD-like program was the equivalent of nearly 320 mhz, since
all processors were just about as busy as they could be. 320mhz computation
10 years ago was a big deal!

And debugging for me was easy: my personal work station only had 4 T8s, but
the executable module was THE SAME as one that could run on 100 T8s. More
processors, more links and each processor got a small set of data to work
on.

Sorry for the spillage - I love Transputers and feel they are an important
piece of computer history in spite of their rather short "mainstream" life.

Gary
Received on Thu Jul 22 1999 - 02:32:35 BST