Despite what it is doing to the rest of the computing industry, the recession seems to be sparing the massively parallel supercomputing market, according to David Moody, European Manager of Intel Corp’s Supercomputer Systems Division. We’re doing brilliantly, he beams. Revenues from Intel’s global supercomputing operation have been growing at a rate of 100% per year, but to what, the company won’t say, and the trend has continued through 1992. Needless to say, the company is also maintaining its lead in the the world’s massively parallel vendor rankings: recent figures from industry analysts InfoCorp show Intel with a healthy 35.11% hold of the market share in 1991, its nearest competitor Thinking Machines Corp with 22.88%, Meiko Scientific Ltd third with 7.15%, nCube Inc fourth with 4.83%, UK-based Active Memory Technology Ltd fifth with 4.58%, MasPar Computer Corp sixth with 4.34%.
Software issues
International Data Corp put the 1990 massively parallel market at $153m and it has been growing at about 20% overall, which implies that the Intel unit had sales of about $65m in 1991 if both the market researchers are right. Such success is attributable to Intel’s many years of experience according to Mr Moody – its supercomputing division was established back in 1984 and now boasts 320 installations worldwide. 1991 was a record year for the company in Europe too where it increased its presence by 16 installations (it now has 50 systems altogether). Its new customers included Grant Tensor Geophysical Corp in England and the French National Office for Aerospace Research and Studies, both of which bought 128-node iPSC/860 systems some of the most powerful in Europe. This spring saw Intel confirming its commitment to Europe with the opening of a European Supercomputer Development Centre in Munich, headed by Dr Thomas Bemmerl, the principal developer of the Tools for Parallel Systems environment. The centre is focussing primarily on software issues, which are considered key to the solving of Grand Challenge applications and for producing successful TeraFLOPS machines. Rather than leading projects, Intel is collaborating with the likes of Daimler-Benz AG, Hitec GmbH, IBM Corp, Inmos International Ltd, Parsytec GmbH and Telefunken GmbH on a number of European Community-funded projects including the European Strategic Programme for Information Technology, Esprit, the Rubbia working group and European Industry Initiative. Work is also progressing well back in the US where Intel is co-funding a research project into TeraFLOPS development with the Defense Advanced Research Projects Agency. Intel has invested around $100m in the scheme – a substantial increase on the $27m Touchstone collaboration from which its current Paragon XP/S massively parallel processor evolved. Apparently, the beta scheduling for this latest Intel offering has suffered a hiccough or two – delivery to the first recipient, Oak Ridge National Laboratory in Tennessee (CI No 1,956) will not now take place until this month. Other customers expecting their machines by the end of this year may have to wait until 1993 too.
By Lynn Stratton
Generally though, progress is considered encouraging. Intel is characteristically coy about releasing figures, but Moody revealed that at least five systems are already on order in Europe (Germany’s largest national research centre, the Research Centre Julich will be the first European recipient) and potential deals in the US have reached double digit figures. Announcements will be made once shipments start. There has been speculation for a while about the future of the 80860 RISC microprocessor, the second generation of which, the 80860XP, is used in the Paragon. It was recently reported in Electronic News that the 80860 was to be de-emphasised and that a switch back to the iAPX-86 architecture, with which Intel’s iPSCs started out, was under consideration. However the company has remained adamant that it will not consider buying in outside produced RISC processors, confirming that it will be using its own in all future p
roducts. We’ll obviously go with which ever processor makes sense, Mr Moody added. Meanwhile, the OEM agreement with Digital Equipment Corp, announced in May (CI No 1,925), is now cast and concrete. Under this, DEC will market smaller to medium range systems based on Intel’s iPSC/860 and recent Paragon technologies. It is expected that DEC will start shipping in early 1993. The two will also be collaborating to convert third-party applications software packages for the Intel architecture, and so to their systems. Their partnership has developed from an earlier agreement to collaborate on a range of massively parallel software products (CI No 1,894) starting with a high performance Fortran compiler. The latter is being designed to accept the language being developed by the High Performance Fortran Forum which comprises Intel, DEC, Cray Research Inc, IBM Corp and Thinking Machines. File management and networking systems will be the likely focus for future projects. DEC’s excursion into the supercomputing world is welcomed by Intel; it is keen to take full advantage of DEC’s distribution network and experience of software, applications, networking and service. DEC’s presence in the smaller to mid-range commercial market also leaves Intel free to concentrate on the high end of the massively parallel computing market. There seems no doubt in the Intel fold about the direction of supercomputing – it is towards massively parallel processing.
Database
Other companies have extolled the virtues of combined vector and parallel processers in ‘hybrid’ configurations, but Mr Moody is convinced that, the days of the vector processor are numbered. He noted also that supercomputing applications, which have previously been oriented towards defence are is becoming more environment driven, being used for research into areas like water pollution, global warming and genetic engineering. Industry is also beginning to catch up with supercomputers, in areas such as fluidics and aerodynamics, pharmaceutical design and even for financial analysis. Although the problem of programming such complex machines has deterred more commercial users in the past, parallel technology seems increasingly to be viewed as a means of harnessing cheap microprocessor power and is proving suitable for applications such as imaging, which requires large amounts of power to create visuals on screen, and for running databases. Oracle Corp has not been slow to get involved with step-company nCube, firmly believing a massively parallel system can run a relational database around 10 to 20 times more cheaply than a conventional mainframe, and that the gap is continuing to widen.