As Digital Equipment Corp storms away from the field and starts scaling the foothills of IBM’s $50,000m annual revenue mountain, Burroughs and Sperry combine to create a $10,000m-a-year business that may still be number two to IBM at the end of the year, and Bull, Honeywell and NEC pool their computer interests outside Japan, all the rules in the general-purpose were changed last year. But Prime, Computer Inc believes thatther will still be a firm place for the comparative minnows that do under $1,000m a year – a figure it just could breach this year – and here Geoff Conrad looks at a couple of the new product lines with which Prime intends to build on its reputation for innovative systems.
Prime Computer UK duly added two new top-end processors to its 50 Series 32-bit superminicomputers earlier this month (CI No 663) – the first of a series of announcements that will include a minisupercomputer from Cydrome Corp and a three dimensional graphics workstation based on the MIPS Computer Reduced Instruction Set microprocessor. The 6350 offers twice the performance – 11.8 MIPS – as Prime’s previous high-end machine; while the tightly-coupled dual processor 6550 is claimed to be rated at 23.6 MIPS, making Prime the only company that can get double performance from dual processors. They won’t say how they do it, only that the two processors are combined in an innovative and balanced relationship. The two processors are symmetric (earlier Prime dual processors were designed in a master-slave relationship, with one processor handling all the input-output and the other working under its direction), both having access to input-output and the system resources. They share the same memory and run under a single copy of the operating system which swaps tasks back and forth as required to balance the load.
It also gives a measure of fault-tolerance in that if one processor fails the other can be restarted to run the applications while repairs are being made. The processors are the first to use Motorola’s new MCA2500 ECL technology – custom-designed VLSI circuits with 2,500 logic gates per chip and are claimed to give extra performance by using ECL level connection techniques off-chip as well as on-chip – earlier ECL circuits were slowed down by using TTL levels off-chip, according to Prime. Each processor has a five-stage pipeline and gains performance from a branch cache: each time the program goes round a loop, the branch cache loads the instructions it chose after the branching last time around into the pipeline it misses in the first and last loop but hits on every other iteration. As most programs loop many times, the savings can be considerable as a miss means emptying the pipeline and reloading with a fresh set of instructions. The system also has a 32Kb cache memory and a segment table lookaside buffer to speed up virtual address translations. The cache and segment table lookaside buffer are both two-way set associative – they are divided into two halves which are addressed simultaneously to improve performance – and they have built-in error correction. Both the 6350 and 6550 support 64Mb of memory and up to 960 users and a new input-output system supports 64 770Mb disk drives for a total of 50 gigabytes. Each input-output processor has an input-output bandwidth of 24Mb per second. Both machines use Prime’s proprietary Primos operating system, with Primix, the company’s version of Unix system V.2 hosted under it. Unix commands can be executed under Primos (with a prefix) and similarly Primos commands can be executed under Primix. Prices for the 6350 – available now – range from UKP439,200 to UKP532,600, while the 6550 dual processor, available in the fourth quarter, ranges in price from UKP643,200 to UKP736,600. Meanwhile at the end of the month, Prime in the UK is set to launch its three dimensional graphics workstation, the PCXL 5500, based on MIPS Computer’s 5 MIPS reduced instruction set processor and developed by Silicon Graphics. Prime claims it was instrumental in bringing the two companies toget
her and encouraged the development of the machine as its entry into the hotly contested workstation market.
Silicon Graphics markets the machine as the Iris-4D and claims a performance of between 5 and 7 MIPS, thanks to the use of 38 custom and semi-custom graphics chips in the design. It runs Prime’s implementation of Unix V.3 with Berkeley 4.3 extensions, beta testing starts next month and shipments in June. It comes with a MIPS floating point co-processor board in a two-tower configuration (the first tower is a 12-slot VME board card cage for all the CPU, graphics and controller boards, while the second includes the power supply and peripheral modules), 4Mb of memory, an Ethernet controller with TCP/IP and network utilities, 170Mb Winchester disk, Unix, C compiler and debugger, graphics library, window manager, eight colour planes with 256 colours, 19 colour monitor with keyboard and mouse all for $74,900 in the US. Other details are not available yet, but the Iris-4D is claimed to perform 140,000 three-dimensional 32-bit floating point transforms per second and render over 4,500 100-pixel polygons per second with smooth shading and hidden surface removal, giving dynamic motion to solid objects. One option includes 24 colour bit-planes, providing more than 16 million colours, four user accessible system planes for overlay and underlay, and a 24-bit Z buffer for more accurate and realistic surface rendering and hidden surface removal. It also offers multi-mode windowing which allows applications using different display modes to run concurrently in different windows: 24-bit colour in one, 12-bit colour in another and so on. It also comes with an optimising Fortran compiler and Sun Microsystem’s Network File System for transparent file sharing in a heterogeneous environment. The Prime machine may not share all these features (it is likely to run the Primos operating system or at least provide access to Primenet so it can communicate with other Prime machines), but as it comes with exactly the same base price, it cannot be significantly different.