Talk:Transputer

From Wikipedia, the free encyclopedia

 This article is within the scope of Computing WikiProject, an attempt to build a comprehensive and detailed guide to computers and computing. If you would like to participate, you can edit the article attached to this page, or visit the project page, where you can join the project and/or contribute to the discussion.
??? This article has not yet received a rating on the quality scale.
??? This article has not yet received an rating on the importance scale.

Contents

[edit] ST20

It's not strictly true to say that the ST20 is no longer built - the cores (c1 and c2) are still being used in devices, like the STi5514, STi5517 etc. They are due to be replaced in new chips, but the old ones will continue being made for quite some time yet. :)

arafel2 10:40 30 Jun 2004 (GMT)


The ST20 was actually a complete redesign. Started in June 1992 and with first silicon in 1994 it combined a lot of features from the T4 series and T9000 transputers. Interrupts and a debug controller were added specifically to support SOC systems and traps and other instructions were added to support operating systems. Two main versions were made. The C4 core was a superscalar design with a Tomosulu style architecture (see patent US5745725 for details) but turned out to be too big for the SOC market at the time. A cut down version (C2) was made shortly after and a minimal form of the T9 instruction grouping technology was added to get back some of the performance.

[edit] The word "Transputer"

message I left on User talk:220.130.0.100: Are you sure that Transputer stands for transmitter computer, not transistor computer? I could only find 2 pages on Google supporting that, one Polish and one Hungarian.. all the English pages say transistor computer. —Stormie 06:29, Jul 23, 2004 (UTC)

Transputer was definitely a combination of 'transitor' and 'computer'. The idea was that the transputer would be sold in a cheap package with just power and link connections and would cost $5. Then you could have a board full of transputers instead of a board full of transistors.

--Robert.Simpson 09:32, 1 Feb 2005 (UTC)

[edit] Ownership of Inmos

The article states "This was too much for INMOS, who didn't have the funding needed to continue development. The company was sold ....".

As I remember press reports at the time, INMOS Limited was owned (wholly? mostly?) by the UK Government, (possibly through the National Enterprise Board?) so it was not the people at Inmos who sold the company.

Songwriter 17:55 3 Jul 2003 (UTC)

Inmos was sold by the government to Thorn EMI in 1984. Having bought it, they didn't really know what to do with the company but kept it going with just enough investment. They sold it to SGS-Thomson (later STMicroelectronics) in around 1989. ST were quite supportive of the T9 development but their focus was on the high volume consumers market, not high end so the T9 was the end of the 'real' transputer development.

[edit] Move this page?

According to the first sentence the all-lowercase "transputer" was the official written form. Shouldn't we therefore move the page to INMOS transputer instead of it being at INMOS Transputer? -- stillnotelf has a talk page 05:10, 23 January 2006 (UTC)

[edit] Request: style correction

In section "Comparison with modern technology" two sentences in a row begin with "Nevertheless", and it looks awkward. Since I'm not a native English speaker, I don't think I'm qualified to tinker with that, so it would be nice if a native speaker takes a look at it and change one "nevertheless" to something else if necessary. --Arny 04:06, 14 February 2006 (UTC)

The first "nevertheless" made no sense in context anyway, so it's gone now. -- stillnotelf has a talk page 05:11, 14 February 2006 (UTC)

[edit] Information and Question

Added something about speed up problems with transputers. Question: HOWTO include T800 block diagram. Have I made an error uploading to wiki commons ? w:de:Benutzer:Lehrig Lehrig 05:23, 5 May 2006 (UTC)

[edit] Clock multiplication

The article correctly states that the external clock was multiplied up (inside the chip). My recollection is that this was a pioneering idea later used on other processors (e.g. the Intel series including the Pentium) to overcome the problem of having a high-speed clock on a circuit board. Today's multi-GHz clock rates would not be possible without this technique. If I am correct it would be good for the article to include discussion of this historic point.Sangwine 11:13, 10 May 2006 (UTC)

[edit] Stack Caching

The article states in a couple of places that the register stack cached the in-memory stack. This may or may not be technically correct, depending on what you mean by caching. On modern CPUs caching means that access to recently accessed memory locations is speeded up transparently to the machine code. This is not the case on the transputers - you must explicitly load values from the memory stack onto the register stack, and explicitly save values from the register stack onto the memory stack. It is true that the register stack was volatile - you didn't know what was on it after certain instructions had finished, and had to reload from the memory stack; in that sense the register stack is a cache. I think this point needs to be clarified in the article in prder to avoid confusing readers not familiar with transputer assembly language. --70.18.247.112 19:28, 31 August 2006 (UTC)

The article states (as of 19th March 2007) "For additional speed the T9000 cached the top 32 locations on the stack, instead of three as in earlier versions." This is wrong; the three deep register stack does not cache memory. The T9000 had a cache which could hold the first few (32???? - I need to check the documentation) locations of the workspace. This was implemented as registers and was fundamental to the way in which the T9000 worked. Roger 21:21, 19 March 2007 (UTC)

[edit] Alice

We could maybe add some reference to the Alice Graph Reduction machine from Imperial College, London, which was a LISP supercomputer that could evaluate multiple parts of the graph in parallel. It represents all that was wonderful and tragic about the transputer. On the one hand, it could do fancy stuff in parallel. On the other, by the time it was finished, Sun workstations could outrun it. Some references to Alice-related publications come up on the web, but mostly to documents behind the ACM paywall.

[edit] Transputers for computer graphics

In the late 1980s, a company called "YARC" produced T800-based co-processor boards for PCs. Each board had 4 T800s with 4 megs of RAM per transputer, and the system had a secondary bus to link the cards together, allowing up to a 16 transputer network. I used a few of these systems with a software called "Digital Arts", which was an implementation of Pixar's Renderman. A scene to be rendered would be chopped up into 32 x 32 pixel "buckets" and the geometry, lighting and textures needed for each would be sent to a transputer, which would send the rendered pixels back to the framebuffer. K8 fan 01:17, 7 May 2007 (UTC)

[edit] The term "Transputer"

As noted above the term "transputer" was meant to refer to the processor as a transistor like system component of a computer. It was also the intention that the term transputer not be a trademark belonging to INMOS but a public domain term for this purpose. It did in fact find its way into dictionaries with this definition as I recall. --Steven Zenith 21:56, 20 October 2007 (UTC)