Any day now, Convex Computer Corp will unveil its fourth quarter 1991 figures, which will reflect a tough year in the minisupercomputer world. After 27 consecutive quarters of growth, Richardson, Texas-based Convex entered the red in its second quarter last year, thereafter reporting a $6m loss for the first nine months of the year. However, according to Convex European managing director John Hughes, analysts have forecast a turnaround in the fourth quarter, with a reduced loss per share to between $0.04 and $0.06 and revenues of $50m, up from $40.5m in the third quarter. In the longer term, analysts are said to be predicting a 1992 turnover of $250m, up from $200m, with earnings per share of $0.60 to $0.80. After a turbulent year, Hughes says Convex’s share price has stabilised at around $13. What has been causing the fluctuation in Convex’s financial performance? A combination of several factors, Hughes maintains: the poor economic climate put pressure on the company’s gross margins and, in the midst of this, Convex was making its product transition to its C-3800 high-end Gallium Arsenide supercomputers. The C-3800 didn’t begin shipping until September, and in the meantime customers were holding tight and not buying Convex’s other wares.

Tea-leaves

With only one C-3800 shipped in the third quarter, the company made a $8.5m loss for the period, while six systems were installed in the final quarter, worth over $2m each, helping to reverse Convex’s previous misfortune. Of course, one other factor that knocked Convex’s shares for six was the recent announcement of rival Cray Research Inc’s alliance with Digital Equipment Corp, whereby DEC will take over all distribution rights to Cray’s $300,000 entry-level Y-MP machine (CI No 1,805). Learning of the negotiations between the two vendors, analysts cautioned that the deal would negatively impact Convex’s business, given that Convex derives much of its revenue from users that are essentially DEC customers. Hughes’ response to such concerns, however, is that the Cray-DEC pact is largely irrelevant. He notes that Cray started to pre-sell its baby system a good five months before it was released, claiming 50 orders over that period. Convex, meanwhile, sold over 200 of its competing machines during the same period. Secondly, Hughes continues, if you read the tea-leaves, you wonder how (Cray’s system) fits into DEC or Cray’s product lines. Why, for example, did Cray give up its own marketing rights to the product, which it didn’t even make itself anyway. Convex’s baby, launched last November (CI No 1,806), is aimed at small server applications, positioned more against Sun and Silicon Graphics servers than against Cray’s offering. Convex’s high-end offerings, meanwhile, come up against Cray’s mid-range offerings in every sale that Convex tries to make, particularly the C-3800 Gallium Arsenide machines; and it was a proud day for Convex last May when the minisuper manufacturer pushed out a Cray X-MP/28 at the University of London Computer Centre and replaced it with a four-processor C-3840 machine. Convex today boasts a total 950 installations, 385 in Europe, where Germany provides a strong source of business across the board with 100 machines, the UK housing 50. This compares with under 100 Cray sales to Europe, though the comparison isn’t really fair, given that most of Cray’s offerings start where Convex’s finish in terms of system size and cost. The Cray Y-MP class Convex machines have one to eight processors using 0.8 micron, 45,000 gate-per-chip GaAs arrays made by Vitesse Semiconductor Corp.

By Sue Norris

The machines, for which a further 10 orders were received during the fourth quarter of 1991, are aimed at oil companies, the mechanical design community, the car industry, computational chemistry companies and researchers, defence projects, and organisations concerned with weather and climate modelling. Hughes says Convex’s reason for being the GaAs guinea pig was that it wanted to present its high-end processing power in an air-cooled system. Only a small pa

rt of the system was fabricated using the GaAs technology, so the impact on the cost of the system has been relatively marginal. The only problems Convex claims to have had with the Gallium Arsenide was a learning curve effect, which led to the gap between announcement and shipment. Hughes maintains that Convex is not interested in the traditional data processing market – the nearest Convex has come to the commercial supercomputing world is through a sale to Dow Jones for market forecasting. Convex’s growth area, according to Hughes, is in file management and file migration. In fact, in a few weeks, Convex will be launching a series of seminars in Europe, on the issue of file serving: Convex has implemented General Atomic Inc’s Unitree file migration software, which now runs on 15 different vendors’ architectures. Convex is already shipping the product, with seven installations to date, mainly in the UK, and three production systems on order. The Texas firm hopes to make some money from providing file services to users of networks incorporating Convex hardware, for automatic file back-up and so on. It will also promote E-Systems Inc’s Emass mass data storage product for supercomputers, compatible with Convex and Cray hardware. In addition, Convex will talk about its own storage product Convex Storage Manager – basically a less sophisticated subset of Unitree or Emass. Hughes is a man with many opinions. Responding to a criticism once made by the now defunct, but one-time competitor of Convex, Floating Point Systems Inc, that Convex, while having a Unixalike operating system, is still basically proprietary (applications such as databases costing as much as $1m to implement), Hughes argues that Convex users have access to the best range of supercomputing applications available – 1,500 altogether. He notes that to continue with the proprietary C-series architecture is essential for Convex’s large installed base, which has large investments in the hardware and software. But Hughes also recognises that there is a clear market for massive parallelism, which needs to be based on an open architecture, around off-the-shelf processors. Convex is known, like Cray, to be involved in research into massively parallel supercomputing, but insists that this technology won’t see a production market until the mid to late 1990s, when a bulk of software is available. Hughes says Convex’s massive parallelism research programme is very software intensive and the company, which is perhaps not as recognised as it should be for its range of application migration tools and compilers, is set to make the first of its announcements on massive parallelism in the first half of this year. Convex will focus on moderately-parallel systems, with 64 to 128 processors, scaling up to a maximum of 1,000. Convex is most successful in the UKP500,000-to-UKP2m price bracket; Hughes says the company’s massively parallel computers are unlikely to exceed UKP4m, the cost of the firm’s most expensive current system. He confirmed that Convex is opting for a MIMD parallel architecture but, except for saying that it won’t be using the Inmos Transputer, Hughes wouldn’t confirm which off-the-shelf processor Convex will use.

Medusa

It is known that the firm is investigating the 64-bit MIPS R4000 RISC in its Medusa project (CI No 1,739), but Hughes says this is only one of a number of research programmes Convex is currently running. Finally, Hughes does not seem concerned about what Cray may or may not do with FPS’ Sparc technology. Like many commentators, he is confused as to what Cray really wants with it, and claims Convex was never interested in making a bid for FPS’ assets, despite the $75m it has in the bank, just sitting there doing nothing. Hughes thinks it’s a mistake to make acquisitions or mergers in the high-technology field, since the companies involved invariably have unique cultures difficult to bring together; Convex reckons Cray is now finding that out for itself.