Hey, fellahs, what did you expect? If you design a horse by committee, you end up with a camel, and that is what the PowerPC triumvirate appeared to find itself with last week – so much so that Frank Soltis’s team up in Rochester has come up with a contingency plan to turn its back on PowerPC altogether and switch to the Intel Corp-Hewlett-Packard Co P7 for future AS/400s. Evidence that all the high-minded design and hardware harmonisation plans have come to naught lies in the fact that inhappy IBMers are leaking like crazy in all directions as they try to pick up the pieces. According to our sister paper Unigram.X and other sources, IBM has given up on the 64-bit PowerPC 620, which has simply proved too buggy and underpowered to meet its requirements for commercial users. Instead, one plan is to use the PowerPC AS chip in commercial RS/6000s, although it will be a variant that adds the memory addressing of the RS/6000, and may, but may not, eliminate AS/400-specific features such as tagged memory. It does include the full PowerPC floating point, but that is weaker than the floating point of the Power series chips, and so something else will be required for technical workstations. IBM’s own Austin unit – rather than Somerset – is expected to supply the company’s own 64-bit part, the so-called PowerPC 630, either in place of, or as successor to, the 620. Meantime, the team in charge of IBM’s parallel RS/6000 SP2s, now part of Mark Bregman’s RS/6000 Division, have also made a unilateral declaration of independence and will come up with separate commercial and technical successors to the current machines, the former using PowerPC 604s in symmetric multiprocessing configurations, the latter using single Power 2 Super chips followed by the 630. Motorola Inc and Compagnie des Machines Bull SA are left to salvage something from the wreckage of the 620 on their own, and that something is being called the 620e. All of which explains why Apple Computer Inc is so keen to hitch its wagon to the BiCMOS star of Exponential Technology Inc: Apple reportedly grew impatient with the internal politics at the jointly-operated Somerset design lab in Austin some time ago and pulled its few people out, keeping them geographically in the area but working on projects far removed from microprocessors. AS/400 watchers – who say Rochester’s got symmetric multiprocessing problems of its own – believe parts of IBM’s 630 may find its way into a 1998 PowerPC AS: higher clock rates and improved (up to 12-way) symmetric multiprocessing scaling will be offered in the meantime, they believe. 1997 will see PCI input-output introduced alongside AS/400’s System Product Division bus on the way to use of the PCI bus throughout, with support for SPD daughterboards, in 1998.
