Now that the dust is beginning to settle after IBM’s Personal System/2 announcement, the Microbytes Daily electronic newswire has been taking a look at the considered reaction of IBM’s competitors and would-be parasites, and, not surprisingly, finds that cloning a PS/2 may not be quite as easy as cloning a standard Personal. The problem, say those who have delved into the machines, is that IBM has distributed logic among several different arrays and chips on the motherboard instead of confining it to a single one. Consequently, it isn’t clear where the individual functions are – in the firmware, software, hardware, or eventually maybe in add-in cards, one industry source commented. Morris Jones, vice-president of technology for Chips & Technologies in Milpitas, California concurs, saying that there is a lot of logic in the chips that isn’t talked about in technical manuals. Jones believes that in order to implement a fully compatible PS/2 system, you are going to have to fully understand that system, stating bluntly that you can’t just blindly reverse-engineer a PS/2 and copy it. He finds that what IBM has in mind for some of the PS/2 functions is not yet clear – we haven’t seen the big picture yet. As an example, he singles out the Microchannel Architecture bus that, unlike that of the IBM Personal bus, allows a co-processor to talk directly to all the peripherals.
This indicates, he speculates, that IBM has put a lot of emphasis on future products and that hardware manufacturers haven’t been told what those products might be or how they might work. Another specialist comments that You have to look at the system as a whole – the DMA, the graphics chip, even the hard disk controller – and know where every gate is and what it does. If a manufacturer finds DMA-related code in two chips, for instance, and does not continue looking elsewhere for additional DMA information, something can be missed, with the result being that the system you design may not work at all. Anybody who doesn’t get down to the last gate is probably missing something, he warns. The implication of this is that IBM has gone to fiendish lengths to make the thing tricky to emulate. How about those who simply want to accept IBM’s open invitation to produce add-ons for the machines? The company insists that it is still committed to open architecture, after all. Well according to Microbytes, makers of add-in boards are also discovering that when it comes to designing products for the PS/2s, all the rules have changed. Not only are input-output ports built directly into the motherboard, raising, in many cases, a question about the need for add-ins at all, but the design of the system itself is making the job more difficult than it has been to build boards for existing PC-DOS boxes.
Some add-on board designers say major issues to be dealt with include mastering the bus, the prevalence of shared interrupts, and how signals are interrupted by boards and the bus. The challenge, according to Richard Rohlf of AST Research in Irvine, California, is having to learn the bus all over again. Matt Zuckerman, president of Advanced Transducer Devices in Sunnyvale, finds the PS/2 systems very sophisticated in the way they integrate the BIOS in the architecture. When the system is turned on, he says, calls are made to the BIOS so that diagnostics can be performed on the various input-output ports. This makes it very difficult, if not impossible, he believes, for standard add-in multifunctional boards to be diagnosed in the same way that the PS/2 performs diagnostics. He also sees the powerful bidirectional parallel port on the PS/2 as another difficult issue for add-in board companies to contend with. Nevertheless Advanced Transducer Devices is one of the handful of companies announcing or showing PS/2 add-ins at the Comdex/Spring show in Atlanta this week. It will offer memory expansion boards using the Expanded Memory System for the Model 30, capable of using 64K or 256K chips and capable of packing up to 2Mb at $10
9 plus the cost of the chips. It will also include RAM disk alternative, print spooler, diagnostic software, and Expanded Memory simulation software, which allows access to multitasking. For the Models 50 and 60, the company has an extended memory multifunction board with 512Kb to 2Mb memory and both serial and parallel ports. To get around the problem with start-up diagnostics, tests aren’t automatically performed on the serial and parallel slots on startup, Zuckerman says. The naked board will be $379, with 500Kb it will be $500 and with a full 2Mb, $970. Both boards will be available late in June, ATD said. Also on the way from the company are a 1,200bps modem card and an AT-to-PS/2 format converter for the PS/2. AST Research has also indicated that it will be showing PS/2 add-in memory cards at the show, including the Advantage/2 for the 50 and 60, which will have optional input-output ports with extended memory from 512Kb to perhaps up to 4Mb although the company was saying last week that the final configuration hasn’t been announced. Microbytes notes that 4Mb would be a considerable accomplishment for AST since it is thought that a PS/2 slot recognises only 2Mb of memory, as evidenced by the 2Mb boards that are the only ones offered by IBM for the machines.
Virtual 386 snag
Meantime the 80386 continues to look a splendid chip for Unix systems, and something of a let-down for those who are addicted to MS-DOS programs. One of the major features of the chip, notes Microbytes, is the virtual 86 mode, which enables users to run several different 8086 programs, each in a different virtual machine, simultaneously on the same computer. A number of companies are working on operating systems that are intended to take advantage of the virtual mode – Locus, Interactive Systems, Alloy Computer Products – and Quarterdeck’s DesqView coupled with its 386 QEMM memory manager software can already take advantage of the 386’s virtual mode to allow multitasking on 386-based AT-alike systems. But benchmark checks of 386-based systems have reportedly revealed severe speed problems with most software written in higher-level languages whenever the computer is operated in the virtual 86 mode. This is because most compilers build code that uses soft interrupts. Although this produces compact code that runs fast on 80286-based machines, and on 80386s in non-virtual mode, the 386 in virtual mode must make a major state change to serve a soft interrupt. Consequently, operations that take a mere dwo clock cycles in non-virtual mode may require from 50 to 80 cycles in virtual mode. Computation-intensive benchmarks indicate that speed degradations can reach 50% or more for certain kinds of programs. As one player commented, It looks like everyone’s got to rewrite the code generators. I’m getting my people started this weekend.