THE TRANSPUTER by Micha Claessen
The age of transputers has started. Atari is almost ready to
launch their transputerboard. Although transputers are not so
new anymore; the industry is using them since a couple of years
already e.g. in mainframes (IBM, Hewlett Packard, Texas
Instruments...). Now there already is a transputerboard for
our everloved ST. It's called KMAX and comes from KUMA. It's a
RISC Processor Development System (they call it that way). It
has a 15 Mhz T414 32 bit transputer from Inmos with 7,5 mips
(million instructions per second). You can own it for £695 (a
price decrease has recently arrived - it costed £995 before
that) which is around 2400 Dutch Guilders (ough!). Therefore,
you get a small yellow case containing a transputerboard also
with 256 Kb ram of memory. The Development System also contains
a Cross-Assembler with an 52 DIN A4 sheets manual. I guess it is
a very interesting Add-on for those who are interested in the
programming of a transputer and also have a lot of money. An
also very interesting computer is the ARCHIMEDES from Acorn.
It's also based on the RISC principle, but it's no transputer.
The fact that this computer has a RISC-(I'll explain 'RISC'
later) design makes it very fast.
The ARCHIMEDES has an 8 Mhz custom designed 32 bit processor:
ARM. It reaches a speed of 4 MIPS... .............(Million
Instructions Per Second) and is about 10 times as fast as our ST
(I'm a bit ashamed to write this down). I played a game called
Zarch on this computer on the HCC days in Utrecht, and I must
admit that I'm heavily impressed about this computer although I
didn't hear any music from it (it has 2*4 sound channels). The
only disadvantage of this computer is that there isn't much
software for it (yet)!
But..you know that Atari doesn't stand still so we'll wait for
the transputerboard from Atari. It's gonna be a transputerboard
containing a T800 transputer from Inmos. This T800 processor is
a successor of the T414 with intern Floating Point Arithmetik
commands and can be used with a clockcycle of 20 Mhz. It can
reach a speed of 15 MIPS and 1.5 MFLOPS (Million Floating Point
Operations Per Second). This is about four times as fast as a
68020 processor. The transputerboard has 5 Mbyte on board (1 Meg
is used for video purposes). The hardware environment is called
Perihelion and Atari hopes to set the standard with this
environment.
It's about time that I try to explain to you the principles of a
transputer and how it works.
A transputer is a processor based on two pricipals:
1) RISC (Reduced Instruction Set Computer)
2) Parallel Processing
First I'll try to explain the RISC principle. There is nothing
mysterious about a RISC design. Most of us already had such a
computer based on an reduced instruction set chip ( C-64, Atari
XL, Apple...). The commands were simple instructions. The Z-80
however was a bit complexer. You could move a whole block of
memory with only one instruction. Therefore it has to be a more
sophisticated and therefore better chip. Also a cokplex command
is e.g a multiply function. The complexer the commands are the
more sophisticated and better the chip has to be.This takes much
space on the chip. The idea behind RISC today: The simpler the
commands the more space you get for other purposes. You can use
the space e.g. for more registers, a cache memory, built in
speed! An other disadvantage of complex processors ( 80386,
68030) is that some instructions are so complex that they
sometimes even don't work properly. This malfuction is called
instruction sensitivity. In a good RISC processor most commmands
take only one clockcycle.
A RISC procesor replaces the complex commands by simple commands
So you can replace such a complex instruction like 'multiply' by
repeating an addition often.
Actually this number of simple instructions is just as fast
as the multiply command and all other commands are even much,
much faster in a RISC chip. Some RISC processors even can run
hundreds of times faster at some instructions than a complex
instruction set computer.
A typical RISC architecture is: 40 commands, 2 adress modes,
almost all commands only need one clockcycle, 150 registers.
The T800 from Inmos that is used in Atari's transputerboard
however isn't really that reduced.I'll talk about the T800 later
in this article.
Now I'll try to explain the Parallel Processing priciple. To
understand this Priciple, you need to know something about the
hardware from Transputers, so I'll first explain something about
the hardware of the T414 from Inmos. This transputer has four
LINKS.These links are used for communication between transputers
and between the transputer and a normal computer. A link is a
bidirectional serial interface. It has a baudrate of maximal 20
MBit/s. Actually a 'normal' computer hasn't such a interface so
Inmos launched the Link-adapter-chips. They can be connected on
the RS-232 interface. The C011 for example converts the serial
link data in a bidirectional parallel interface with two
statusregisters.
Also,a Decoder PAL and a Control PAL are used that are important
for communication between computer and transputer. Further,
the Transputer has Memory Managment but that isn't that
important for understanding parallel processing (I must control
myself for talking about things that you're not interested in).
The T-800 also has a very fast 4 Kbyte cache memory in its chip.
Imagine that you call a program that runs on a transputer; a
process modell. You can see each part of it as a process. Each
process can communicate with an other process (with the links).
By a command you can start a process and when the process ends,
it sends back information.Naturally, a process can start another
command and a group of commands can be seen as another process.
Processes may be assigned either high or low priority, and there
may be any number of each.
Maybe you know what RAY TRACING is. This is a technique which
can generate very realistic images but requires massive amounts
of computer power. This is an ideal application for transputers
as the calculations for each pixel (picture element) on the
screen are indepentent of one another and so can be done
in parallel on seperate processors.
So parallel processing on just one transputer isn't really
interesting but if you switch some transputers together.........
you get power which' limit lies maybe in the bandwith of our
senses (intellectual joke isn't it? ). You know that the T-800
has 4 Kbytes Ram on chip so you can split a programm, send the
parts to the transputers (memory used as a buffer) and let them
do the job! This is of course a really very short intruduction
to parallel processing, but who knows I'll explain a bit more
some time.
I'd now like to tell more about the T-800 from INMOS, because
ATARI is gonna use it in their transputerboard. The T-800
transputer is a 32 bit CMOS microcomputer with a 64 bit
floating point unit and graphics support. It has a 4 Kbytes on-
chip RAM for high speed processing, a configurable memory
interface and four standard INMOS communication links. The
processor speed of a device can be pin-selected in stages from
17.5 Mhz up to the maximum allowed for the part. A device
running at 30 Mhz achieves an instruction throughput of 15 MIPS.
The 64 bit floating point unit provides single and double length
operation to the ANSI-IEEE 754-1985 standard for floating point
arithmetic. (the same as the 68881 ) 1.5 Mflops at 20 Mhz and
2.25 Mflops at 30 Mhz (The cray II only has 2000 Mflops !!!!!!!)
High performance graphics support is provided by microcoded
block move instructions which operate at the speed of memory.
The T-800 can directly access a linear address space of 4Gbytes.
The T-800 is also pin compatible with the T414 so the T-800
can be plugged directly into a circuit designed for a 20 Mhz
version of the T-414. The instruction set achieves efficient
implementation of high level languages and provides direct
support for the occam model of concurrency when using either a
single transputer or a network (Occam is a specially designed
programming language for transputers that supports parallel
processing and is designed by INMOS - Kuma also launches a
package called K-Occam at £139.55). Maybe, some day, I'll show
you an example of OCCAM.
I'd now like to compare some well known processors with the
T-800.
Processor: Whetstones/second single length:
Intel 80286/80287 8 Mhz 300K
IMS T414-20 20 Mhz 663K
NS 32332-32081 15 Mhz 728K
MC 68020/68881 16/12 Mhz 755K
ATT 32000/32100 1000K
Fairchild Clipper 33 Mhz 2220K
IMS T_800-20 20 Mhz 4000K
IMS T_800-30 30 Mhz 6000K
So much for this time!
If you have the need to call me (for some unknown reason) or you
want to send me a new year's card I give you my address and my
telephone number.
Micha Claessen
Winterstraat 1
NL-6086 ED Neer
tel. 04759-1270
Disclaimer
The text of the articles is identical to the originals like they appeared
in old ST NEWS issues. Please take into consideration that the author(s)
was (were) a lot younger and less responsible back then. So bad jokes,
bad English, youthful arrogance, insults, bravura, over-crediting and
tastelessness should be taken with at least a grain of salt. Any contact
and/or payment information, as well as deadlines/release dates of any
kind should be regarded as outdated. Due to the fact that these pages are
not actually contained in an Atari executable here, references to scroll
texts, featured demo screens and hidden articles may also be irrelevant.