PMAX (DS3100/DS2100) Specifications sheet

A quick blurb on DS2100/3100 specs

Runs at 12 (PMIN/DECstation 2100) or 16.67 (PMAX/DECstation 3100) MHz, 83/60ns Cycle Time
Scalar execution completes one instruction every cycle, classic 5 stage single pipeline.
Separate 64K Instruction and 64K Data caches - 128KB total 1 cycle latency storage

implemented with CY7C166 16Kx4, 20ns static RAM (16.6MHz PMAX)
both caches are direct mapped, 16K entries
4 byte line size, since that's the usual word size and no sense to make any bigger as R2000 only has 32 bit data bus)
16 bit tags, also implemented with the CYC7C166's
write through cache policy
Write Buffer chips (R2020s) helps the write latency a bit without implementing write back cache.

DMA is NOT implemented - all data transfer ops must be load/store through CPU - thus making it very slow for a lot of operations!

uses proprietary 80 pin Double sided SIMM that holds 2MB plus 2 bits parity (18 bit width total), each (not ECC)
Each piece utilizes 18 TMS4C1024DJ, uses 9/side, 60/70 ns
has one NS 1926876 01 per side
side 1 markings: 5019464-02 b1p1
side 2 markings: 5730735-01 29089 8398-03 24989
5 Cycle DRAM stall penalty for each cache miss (ouch)

Frame buffer socket, supports PMAX-CFB(VFB02, color, 1MB), or PMAX-MFB (VFB01, mono, 64KB). Both support 1024x864 resolution. PMAX-CFB is 256 color pseudocolor (mapped color palette), and PMAX-MFB is black and white only - does *not* support grayscale. Of course, the CFB can show shades of gray on a color monitor, but why?

These boards are "dumb", mostly just memory and can only hold one frame, hence the name "frame buffer".
Graphics is unaccelerated! Just about all graphics ops are done through the CPU, which makes it really slow.
Supports a single 16x16 2-color hardware graphics cursor (for mouse pointer). This is done in hardware and does not need the CPU to draw. Of course, mouse location and programming mouse shape still needs to be done through the CPU.
The frame buffer board fits in the big socket parallel to, and near the main memory SIMM slots. See photo. Looks like a huge oversized SIMM.
RS343A/RS170 RGB Color or grayscale (w/1MB video memory) or B&W (w/64K video memory) monitor output

Max transfer rate I've seen on my pmax is around 490KB/sec /dev/zero to network, and 450KB/sec from network to /dev/null. These tests were done through a Fast Ethernet switch to a machine known to be able to saturate 10baseT Ethernet. This is limited by the CPU, so I imagine the pmin is slower.

Max transfer rate I've seen is around 960KB/sec from /dev/zero to disk, and 895KB/sec from disk to /dev/null.
Note you must have a terminator on the external SCSI connector for normal operation. It may well work fine without it - my system does - but your data integrity may suffer. I have a jury-rigged terminator (I don't have the official terminator but hook up an external Centronics-50 terminator to the external 68-to-50 cable.

Mouse and Keyboard ports (mouse or a mouse terminator *REQUIRED* for boot)
Two DECnet (MMJ 6P6) connectors for modem and printer, can be used for serial console
Ultrix Operating System (No less than 12 M RAM must be in the machine for 4.2!)
- Latest version: 4.5, most well known: 4.2a, 4.4
- Does not support shared/dynamically linked libraries! (which means, huge binaries, wasted memory, and nonupgradeable C runtime!)
- X11R6 available.... Xws (R5) and Xcfb (R6) (different for different versions of Ultrix)
- Based off of AT&T code, so licensing ... SCO???! Bleah...
NetBSD was ported to it, and it should be the OS of choice today!

Supports Shared Libaries, ELF binary format, less disk/io intensive, faster than Ultrix for most applications as disk i/o is bottleneck. Compile binaries static to revert to Ultrix-like speeds.
Uses X11R6.3 with shared libraries
NetBSD is freely available by BSD license at http://www.netbsd.org
There is a Linux port, but it's not very functional for this machine. Specifically, unless someone writes code, there's no console nor scsi. Check out Karel's PMAX Linux page for more details.