Hottest Atari around

Press, 6 October 1987, Page 51

Hottest Atari around

By

GARETH POWELL

The star of the recent London Microcomputer Show was a computer that was not on display. It was discussed, talked about, rumoured. I heard more lies about it in four days then I ever have in a month of Welsh Sundays. Several friends of mine in the trade, who should know better, averred it did not truly exist. They were wrong. The computer is an Atari ST, driven by a transputer. It is beyond debate the fastest personal computer in the history of the world. And it has the capability of getting faster, very much faster, with the greatest of ease. How fast is fast? In the real world in which we all live, standards mean little and frequently mislead. Instead, let us take a practical test set up specifically to show the differences.

A major financial modelling spreadsheet was constructed to try to forecast, using every sophisticated mathematical forecasting program in the statistician’s armory, where the United States economy will be in 10 years.

For such figures to be produced, an immense amount of number crunching is needed. On a standard IBM PC AT, which is a very quick machine, the computation took a tad over 46 minutes.

On this new Atari computer, with its transputer engine, it took just on eight seconds. And with a bit of tweaking, it was able to bring the time down to four seconds. How was this achieved?

Mainly through the use of a totally new. type of chip called a transputer, developed at Inmos, an English company that has had its ups and downs and is now, effectively, government owned. The transputer chip is a 32-bit RISC powerhouse. The 32-bit part is relatively easy to understand. The processing is carried out and the information churned around within the chip on a super highway 32 lanes wide. The RISC part is something else. As computers have developed, the temptation has been to fill the central processing units — the heart of the computer — with more and more instructions. These have now become so complex they slow down the working of the chip. But some of these instructions are merely sophisticated improvements on other instructions. Multiply is, in essence, exactly the same as adding a figure many times. With a RISC chip (the letters standing for reduced instruction set), you sacrifice sophistication for speed and move the essential smarts either into other chips or on to software.

But the transputer, although it is a true 32-bit RISC chip, goes much further than this. Besides having all the standard functions of, say, the Intel 80386, it also has a built-on maths co-pressor which, because of its

physical closeness, works far faster than a separate chip. It also has built into the chip a substantial dynamic random access memory — four megabytes which can be increased to 64 megabytes when the occasion should arise.

Finally, it has a communications facility to connect it with other transputers so that tasks can be apportioned between the chips available, with a great increase in speed. This parallel processing has been an area of exploration for many companies, which .are interested in more speed, more power. The Inmos transputer is constructed specifically for that task and there appears to be no limit to the number that can be strung together.

I worked with an experimental computer with 20 of these transputers interconnected. The Inmos boffin who, probably quite illegally, demonstrated it and let me have a play complained bitterly because they will not let him hang on another 20. This would give him the computational speed and power of several DEC VAX stuck together. Even before you start chaining these transputers together, they are chips of quite awesome power. Whetstones come from a Fortran program which is specifically designed to test scientific applications. Take as a baseline the current ultra-fast chip which is the true 32-bit Motorola 68020. Run it at its highest clock speed of 25 MHz. Give it a maths co-processor for support, the Motorola 688881. Then you have an unbelievably fast machine which will scream along at 1500 Whetstones a second.

Now take a single Inmos TBOO transputer, link it to no other chip and then apply the same test. The result is 4000 Whetstones.

Put it another way. The current Inmos transputer you can buy off the shelf works at- 20MHz and runs at 10MIPS. As in millions of instructions per second. The new version, the one I was playing with that will be released to the public next March, works at 30 MHz and runs at 15 million MIPS. This is more than three times as fast as anything else on the market, and 10 to 20 times as fast as most. We are still only talking of a single chip, not parallel processing. • Atari will officially announce a transputerdriven machine at the Comdex show which is only weeks away. • Atari will demonstrate the machine at the Hanover trade fair in March.

• Atari will place a basic Atari ST on sale next July. It will contain one transputer and work at 15 MIPS.

• Another version using four transputers, which fit in the same space as a single Intel 80386 chip, will be shown before the end of next year. These computers will be targeted initially at the educational and scientific markets and then, later, to the public.