home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
No Fragments Archive 10: Diskmags
/
nf_archive_10.iso
/
MAGS
/
INFO
/
IINFO53.MSA
/
TEXT_ATW.TXT
< prev
next >
Wrap
Text File
|
1991-07-11
|
7KB
|
124 lines
From ST REPORT
> ATW & Inmos STR FOCUS A Comprehensive look at the ATW's future
=====================
H1 INMOS TRANSPUTER CHIP AND THE RACE FOR TERA FLOPS
====================================================
by Mike Stepansky
Has the Atari Transputer Workstation (ATW) been discontinued
by the Atari U.K. officials? I think this rumor is somewhat over-
exaggerated, but I do know that transputer technology is NOT dead
in the private sectors and in certain universities. Also, I was
not surprised to learn that most transputer scientists argue that
parallel-processing computer will be and should be the next wave
of computer technology after the today's limited processors
architecture, like those current RISC or CISC architecture
implementations.
Anyway, I have been told that this fall will mark the 4th
anniversary of the introduction of the INMOS T-800 Transputer
microprocessor. In the first few years the T800 had by far out
performed its competition such as Intels' 80386 and Motorola's
68020. Even in conjunction with their corresponding floating
point coprocessors such as the 80387 or 68881, the competition
could not surpass the performance of the T800 with its on-chip
floating point unit. In the meantime, the competition has not
slept. More and more powerful devices have been introduced to the
market. At times, it appears that the microprocessor development
groups are coming out with new products even faster than Detroit's
three giants issue their new car models.
In traditional computer technologies, as we all know,
performance increases are gained by faster and faster CPU's. This
puts microprocessor vendors under pressure to provide the market
in very short development cycles with new and faster devices which
quite often are not completely designed or show only minor
improvements (a gain in performance by a factor 2). On the other
hand, Transputer based parallel processor technology is inherently
scalable so that additional performance can be gained by
increasing the number of processors and not by waiting for the
next generation. Although the T800 has been in production for
almost 4 years, T800 systems have been able to grow with
performance requirements up to the present and most likely could
continue do so.
If the competition does not sleep neither does Inmos.
Without the pressure of issuing a minor enhancement each year, the
Transputer developers at Inmos have used the time to make real
improvements. Basically, 2 major advantages are realized with the
H1 (codename for the latest Inmos Transputer chip):
1) A 10 fold increase of processor and communication links
performance. This results in a 20 MFLOPS and 150 MIPS, thus
giving you a total data throughput of 80 megabytes per second.
Think about it! This is now available on a SINGLE chip! This
puts an end to those worrisomes when you need higher clock
frequency in today's serial processors (Intel or Motorola), yet
where the law of quantum physics intervene.
2) Also, there is a modified channel communication protocol
in conjunction with an on-chip Virtual Channel Processor (VCP) and
a new switching chip (called C104) allowing each processor to send
messages directly to any other processor. This is done without
requiring any intermediary processors between sender and reciever
(wormhole routing). Every processor can be every other processors
direct neighbor. Hypercube fashion is no problem at all!
In addition to these improvements, the H1 will be binary code
compatible with the T800 so that software developed on the Txxx
series can be directly transferred to an H1 system. The result of
this will be that H1 systems will have far better software support
with development environments and application software than any
other device of its performance class which is used in MIMD
machines. Also, there will be special hardware components
available which can interface between the Transputer links from
the Txxx seres and an H1 link. Specifically these features allow
the integration of the performance enhancements of the H1
transputer directly in existing Transputer environments. All
previously developed (or purchased) Transputer host system
interfaces or Transputer I/O modules can be used in conjunction
with H1 components.
"So...What is the name of the company that sells and develops
the state-of-the-art transputer?" you might ask. Unfortunately,
not just Atari U.K., but Parsytec, as I have been told, is an
independent system manufacturer located in West Chicago, Illinois.
Parsytec has chosen the microprocessor it will use for their
future systems very carefully. Parsytec has decided to stay with
Inmos in developing the next generation of distributed memory
processing systems. The European Economic Community has awarded
Parsytec, together with 4 other partners, a contract to design a
General Purpose MIMD Computer based on the H1 microprocessor
device. It is interesting to note that Parsytec informed me that
their partner in U.K. already bought one or two ATW's and planned
on improving it to H1 transputer chip in Europe!
Their goal is "to develop an effective architecture for
general purpose parallel computers." Additionally, Parsytec's
trust in the new H1 is so strong that they have decided to lead
the European Teraflop Initiative (ETI) with a proposal for a 2^16
(65,536) H1 system with a scheduled delivery date late in 1993.
Let's hope if they win the 1.6 TERAFLOPS race in their H1 system!
Well, to close this statement, I am still looking forward to
Atari U.S. to integrate this H1 Inmos system in their future
system, if possible and by all means necessary to save Atari
position financially in the 1990s. By the way, news just in from
Inmos: H1 has just been named "T9000"! Welcome to the true 32-
bit parallel computer world! Is ATW dead? I doubt it and the
price tag continues to drop, thus making today's RISC and CISC
chip look outdated, depending on applications.
<<<<*>>>><<<<*>>>><<<<*>>>>