New midrange members of the Hewlett-Packard Precision Architecture Computer Family - includes related article on double-sided surface mount process - technical

Hewlett-Packard Journal, June, 1989 by Thomas O. Meyer, Russell C. Brockmann, Jeffrey G. Hargis, John Keller, Floyd E. Moore

New Midrange Members of the Hewlett-Packard Precision Architecture Computer Family

NEW MIDRANGE HP PRECISION ARCHITECTURE computer systems have been added to the HP 9000 and HP 3000 Computer families. The HP 9000 Model 835 technical computer and the HP 3000 Series 935 commercial computer share the same system processing unit (SPU). Designed with significantly improved floating-point and integer performance, the Model 835/Series 935 SPU meets the computational needs of mechanical and electrical computer-aided engineering (CAE) and multiuser technical and commercial applications.

The HP 3000 Series 935 (Fig. 1) is configured for business applications and runs HP's proprietary commercial operating system, MPE XL. HP 9000 Model 835 products include the Models 835S and 835SE general-purpose multiuser computers, the Models 835CHX and 835SRX engineering workstations with 2D and 3D (respectively) interactive graphics, and the powerful Model 835 TurboSRX 3D solid-rendering graphics superworkstation with animation capability (Fig. 2). All Model 835 systems run the HP-UX operating system. As a member of the HP Precision Architecture family, the Model 835/Series 935 SPU supports a wide variety of peripherals, languages, networks, and applications programs.

User Requirements

Like its predecessor, the Model 825/Series 925 SPU, the Model 835/Series 935 SPU's definition was driven by requirements from several different application areas. In addition to the requirements of small size, low power dissipation, low audible noise, flexible I/O configurations, and tolerance of a wide range of environmental conditions normally required for a midrange technical or commercial product, the Model 835/Series 935 SPU design addresses several other needs. For scientific computation and mechanical and electrical CAE applications, high floating-point and integer computational performance is desired. The Model 835/Series SPU provides more than a 300% increase in floating-point performance and more than a 50% increase in integer performance over the Model 825/Series 925. The Model 835/Series 935 has been benchmarked at 14 MIPS and 2.02 MFLOPS.

Customers who own or plan to purchase a Model 825 or Series 925 want the ability to upgrade to the faster Model 835/Series 935 without having to replace the whole computer. To meet this requirement, the Model 835/Series 935 processor is designed so that an upgrade can be easily done at a customer's site by exchanging two boards. Because of HP Precision Architecture compatibility, user applications migrate and realize enhanced performance without modification or recompilation.

For all application areas, main memory capacity is an important influence on overall system throughput. To meet increased memory requirements, a compact, double-sided surface mount 16M-byte memory board has been made available. Designed to work in any of the Model 825/Series 925 or Model 835/Series 935 products, this board doubles memory capacity to either 96M or 112M bytes depending on the configuration.

Design Overview

The Model 835/Series 935 uses many of the same components as the Model 825/Series 925 SPU. Common components include the mechanical package, the power supply, I/O cards, the I/O expander, the battery backup unit, and the memory cards. This high degree of commonality not only assures easy upgrade potential but also minimized design time.

A block diagram of the Model 835/Series 935 SPU is shown in Fig. 3. The two boards unique to this SPU are the processor and processor dependent hardware (PDH) boards highlighted in the block diagram and shown in Fig. 4.

The following sections will explain the approaches taken to meet the performance requirements mentioned earlier. In addition, the design considerations for a compact 16M-byte memory board using a new double-sided surface mount manufacturing process will be discussed.

Processor Board

The Model 835/Series 935 processor board reuses much of the technology developed for the Model 825/Series 925, a practice frequently called "leverage" within HP. Eight VLSI integrated circuits make up the core of the processor board: the CPU (central processing until), the SIU (system interface unit), two CCUs (cache controller units), the TCU (TLB controller unit), the FPC (floating-point controllers), and two commercially available floating-point chips. Of these, the CPU, SIU, TCU, and two CCUs are functionally identical to those used in the Model 825/Series 925 processor but run 20% faster. These parts were designed in HP's NMOS-III VLSI process. The FPC and the floating-point chips, new for the Model 835/Series 935 processor, will be discussed later.

In addition to faster VLSI, a number of performance enhancements over the Model 825/Series 925 processor board are found on the Model 835/Series 935 processor board. These include:

* An eight-times-larger cache (128K bytes by 2 sets, unified instructions and data).

* A two-times-larger translation lookaside buffer or TLB (2K instruction entries and 2K data entries). Since HP Precision Architecture defines page sizes to be 2K bytes, this allows 8, bytes of main memory to be mapped directly into the TLB.