Although 9672-R CMOS small mainframes are barely getting into the volume production state – we expect IBM to have shipped only a couple hundred by the end of last year – the impressive qualities of this new processor have stunned the mainframe market. Not since 1979, when the 4300s were announced, has an IBM mainframe had so telling an impact on older machines. In some ways the 9672-R is less radical than the 4300. In terms of hardware pricing, it costs just over half as much as the small 511-series 9121s it can replace. The 4300 was a quarter to a third the cost of predecessor 370s. But the 9672-R is not only an alternative to the low end of an earlier series of machines. It is also a powerfully attractive replacement for 3090s of under 60 MIPS, all 3080s and 4300s and a potential rival of small 9121s. Another difference between the 9672 and the 4300 series – one that may prove far more important – is the way it will grow. The 9672 will, according to IBM, have engines with about twice the current power by 1996 and even faster circuitry not long after that.
Gruelling
This kind of progress is necessary if IBM wants to keep the mainframe in the fast, gruelling race against alternative kinds of systems. Although the 4300 provided a tremendous price-performance improvement over the 370s that preceded it, it was by any measure a conventional IBM mainframe. Users who flocked to the 4300 for economic reasons expected IBM eventually to offer some moderately faster but otherwise similar models, and they were not disappointed. IBM built the 4331 into the 4361; it replaced the 4341 with the 4381. The 4300 line caused a financial problem for IBM. Customers took the 4300 announcement as a signal that IBM would soon replace the larger machines in the 303X group with processors that were as greatly improved as the ones at the low end. Sales of 303X machines quickly dried up; users instead acquired 303X machines under IBM’s monthly rental plan. IBM’s income from mainframes was thereby deferred and Big Blue reported some disappointing financial results. The mess was finall y cleared up with the announcement of the 3081-D, which was quite late compared with IBM’s original intentions. The effect of the 9672-R on IBM’s large mainframe business isn’t yet visible. IBM’s sales of big water-cooled 9021 processors were up last year compared with 1993. The computers that the 9672 will most directly affect, the 9121, are also doing well. But the current situation is illusory.
By Hesh Wiener
IBM’s mainframe revenue is considerably below the level of a few years ago and even the current upturn in demand is not likely to yield an increase in year-to-year revenue because of unit price cuts. If there is any factor cushioning the impact of the 9672 on 9X21 machines it is the lack of software for systems with more than six engines. But in computing, individual belated events usually don’t change trends. When the full collection of systems software for big 9672s is released and later, when the S/390 CMOS chip set enters its second generation, owners of large- engine mainframes will be in for a shock. The market will catch up all at once, re-evaluating all IBM (and compatible) systems in comparison to the price and performance of 9672s. The 9672-R machines, with a single central electronic complex, are slated to rise to more than 100 MIPS of power by 1996; the second generation of 9672-E models could provide more than 750 MIPS. Thus the CMOS computers will span the entire range of old IBM mainframes and extend the scope of the systems very considerably. Along the way, the price of a MIPS of CMOS 390 power will fall and so will the cost of maintenance. Barring a change in the competitive picture, IBM will probably increase service costs for its older machines once or twice by 1996. Inflation will also favour the newer machines over older ones; electricity and, in most locations, physical plant costs will rise. The investments users (and lessors) are making in 9021 equipment (and to a lesser degree, in 9121s) are going
to evaporate. If IBM has a large enough amount of money tied up in its lease portfolio, it could offer upgrade deals to preserve the residual values of some machines – 511 and 711 class processors, for example. But it would be unwise for anyone (outside of IBM, that is) to bet that this will happen. We believe users should consider selling all the mainframe hardware they own and then leasing back whatever they need on as short a basis as they can arrange, with 24 to 30 months about the maximum. Because most of the original 9672s will be upgraded to faster models, leasing might be a good choice for those machines, too. IBM is offering attractive bundles of hardware, software, maintenance and financing on 9672s. But users should insist on having the costs broken out to determine if a third party lease might be cheaper, more flexible or both.
Risks
It is our impression that big companies, particularly those who get some coaching from one of the major third party lessors, have an easier time getting IBM to co-operate than smaller shops. IBM apparently will not provide an accurate price breakdown unless pressed. The bundled deals benefit IBM, which can book the total amount as a sale and thereby boost its apparent revenue and profit in the quarter the deal is signed. But we don’t think the bundled arrangement will be good for every lessee. If a lessee in a bundled deal must sell, sublease or otherwise dispose of a 9672, some of the hardware cost might be recovered. But the prepaid software and maintenance fees could be lost. We don’t think IBM is making the 9672 bundles cheap enough to offset the risks to the lessee. At the very least, users who find that IBM is offering the most economical deal should study a cost breakdown and consider financing the hardware under a separate contract.
From the December 1994 edition of Infoperspectives International, published by Technology News Ltd, (C) 1994 Technology News Ltd.
