Apollo DN10000: Quad CPU/128Mb RAM workstation from 1988 [pdf]

Original Article ↗ Hacker News Discussion ↗

Hardware, Specs, and Cost

  • DN10000 supported up to four CPUs and 128 MB RAM in 1988, which was exceptional even a decade later.
  • Maxed‑out systems reportedly cost around $250k then (~$660–700k in today’s money, though exact inflation equivalence is debated).
  • Some commenters stress that unlike many “supports up to” claims, fully loaded 128 MB, multi‑CPU units were actually sold and deployed.
  • Design included both VME and ISA slots, plus serious attention to thermal design and office‑acceptable noise.

Performance and CPU Architecture

  • Apollo’s PRISM CPU was a 64‑bit, VLIW‑like design; some see it as an ambitious early RISC/VLIW experiment later influencing PA‑RISC and Itanium.
  • Estimates in the thread peg PRISM’s MIPS roughly comparable to a later 486DX2/66.
  • There is disagreement on whether 486‑based servers were “dramatically” slower; consensus is that various RISC workstation lines eventually lost out to commodity CPUs.
  • One commenter with VLIW hardware experience says virtualization is technically feasible and cache behavior is not inherently worse than superscalar out‑of‑order CPUs.

Domain/OS, Unix Personalities, and UX

  • Domain/OS is remembered as powerful and innovative (versioned filesystem, distributed FS, diskless clients, strong networking, multiple Unix “personalities”).
  • The dual BSD/SysV personality mechanism used executable “stamps” and environment‑dependent path resolution to choose the right userland and syscall semantics per process.
  • Others recall it as complex, awkward for C development compared to BSD‑derived systems, with sockets feeling bolted on and Internet support de‑emphasized.
  • The graphical environment (Display Manager) is described as extremely capable for power users but confusing and “weird” for newcomers.

Real‑World Use and Applications

  • Widely used in the late 80s–90s for CAD, PCB design, SPICE simulation, VLSI tools, configuration management, and as file/compute servers in universities and labs.
  • Examples include FAA command centers, UK government‑funded university CAD labs, and campus‑wide home directory servers.

Graphics, Input Devices, and Industrial Design

  • Graphics hardware (especially DN10000VS) is remembered as state‑of‑the‑art: 3D, Z‑buffer, antialiasing, texture mapping, 40bpp, and high‑resolution displays.
  • Users fondly recall laser mice with mirrored pads, spaceball 3D input, dual framebuffers with alpha/Z for interactive graphics and even hypothetical gaming.

Emulation and Preservation

  • MAME now emulates several Apollo systems (e.g., DN3500). Domain/OS install images are available on archival sites, making historical exploration feasible, though resource‑intensive.

Historical Context, Pace of Change, and Comparisons

  • Multiple comments contrast 1988–1998’s rapid hardware evolution (68k → 486/Pentium, Alpha, multi‑CPU hobby systems) with what they perceive as slower visible change post‑2010.
  • Others note exceptional mid‑90s machines (e.g., Macs and Amigas with unusually high RAM ceilings) but emphasize that broad multi‑CPU, 128 MB workstations were rare.
  • Some extrapolate to modern dual‑socket Epyc systems (hundreds of cores, terabytes of RAM) and wonder how quaint today’s AI clusters will look in 30 years.

Marketing Materials and Aesthetics

  • The brochure’s hand‑drawn watercolors, diagrams, and logo draw a lot of admiration.
  • Commenters miss this kind of lavish long‑form, print‑style technical marketing compared to today’s more generic web pages.
  • Early CG demo films rendered on Apollos are cited as part of the era’s HPC marketing culture.

Units, Precision, and Price Comparisons

  • Several people object to highly precise inflation conversions (e.g., “$663,937.02”) as “false precision,” arguing that only a rough one‑digit estimate is meaningful.
  • Similar annoyance is expressed about over‑precise metric/imperial conversions and sloppy use of Mb/MB/MiB, especially when both storage and network bandwidth are discussed.
  • There is debate on how to meaningfully compare historical system prices at all: CPI vs IT‑specific indices vs qualitative changes in what computers enable.