[ Date Index ] [ Thread Index ] [ <= Previous by date / thread ] [ Next by date / thread => ]
On Thu, 25 Oct 2012, tom wrote:
I did some occam work on their PC simulator. Once I walked down a corridor and found, next to four lifts and no humans, a large black filing cabinet thingy with 1024 transputers in it - a computing surface I think they called it. Probably one the first ever built.Just sitting there with no-one around.
The Computing Surface was Meikos name for their product, or just a "Surface". The large black boxes were in-fact 19" racks with nice outer skins on them. They would take what was called an M40 - meaning it could take 40 boards. Each board was about 14" square. Meiko went for function rather than transputer density, so the most transputers they ever had on the board was 4 (usually with 4MB of RAM, although as memory density increased it went up to 16MB per cpu). The reason for only 4 was the backplane which carried the switch network - no more hand-wiring up the transputers - Meiko had developed their own switch chip (ahead of inmos) as well as a supervisory bus which allowed control of all the devices - reset, some rudimentary monitoring and programming the switch chips...
(a transputer had 4 links, so 4 on a board is 16-links per slot, so hand-wiring it gets complex quickly)
So the max possible in a single Surface was 40 x 4 = 160 transputers.I think the biggest one they ever sold was to Toyota who were using it for some very advanced (for the time) raytracing to generate the advertising material for their cars (or mixing the paint or something like that). That was 3 or 4 M40's linked together (there was a 16-link plug-in card)
There were many plug-in cards - e.g. graphics - the 2nd gen card which I wrote all the low-level drivers for could work in banks of 3 - one each for R,G and B.
Do you remember way back, ITN's news at 10 where they soomed over a night-time London and up to big ben? That was rendered on a surface - the chap who did it did 30 different renders for each minute past the hour in-case they were running late...
There were other boards - One had a SCSI & Ethernet interface - I worked on that and Minix was ported to it - 8MB of RAM, no MMU and it ran Minix... However we quickly replaced that as it really wasn't viable.
But people wanted faster & inmos wasn't delivering, so they moved on to biards with 2 x i860's on-board and 4 transputers - leaving the transpuiter (2 for each i860) as nothing more than comms chips (shared memory) then they had a sparc board as the host board doing away with the need to front-them with a PC/VAX or Sun. A few other specialist boards were made - e.g. a high speed parallel interface to read Ampex video/data tapes (synthetic apperature radar imaging)
But even the i860's weren't fast enough, so the next gen. the CS2 - in the picture I posted earlier was entirely sparc based.
I worked mostly on the low-level stuff, doing all the test/diagnostic/board bring-up code when I then turned into the device drivers, I didn't do too much applications stuff.
When I read how Zaphod felt about just having to steal the heart of gold a few years later I knew exactly how he felt but without the built in getaway vehicle. I think I spent a good 10 minutes going through the possibilities before carrying on as normal.
I guess being surrounded by it for so-long sort of makes me feel a little "meh" about it all now...
Gordon -- The Mailing List for the Devon & Cornwall LUG http://mailman.dclug.org.uk/listinfo/list FAQ: http://www.dcglug.org.uk/listfaq