ICL 1900 Series Computers
Processors
There were five series of 1900 processors, each series being an improvement on technology and speed from the previous. The initial series were a number without any letter suffix, the next series had the E & F suffixes, the third series had an A suffix, the fourth series had an S suffix and the final series a T suffix. The FP6000, from which the series originated, became the 1904. The machines were made at two factories, Stevenage (the smaller machines) and West Gorton (the large machines). The architecture was different between the two factories, but all the machines gave a common program interface, differences being hidden by the EXECutive. The large machines did have a couple of additional instructions available (MVCH & SMO). A hardware Floating Point unit was an optional extra (not normally required for commercial type work) as was a real-time clock. Similarly, some extracodes and facilities (Sub-Programming & Trusted) were not available on all EXECutives. Most machines, via EXECutive, supported multi-programming, although the smaller were single or dual programming. Unlike IBM machines, the systems were not partitioned; all resources (memory and peripherals) were held in a common pool and allocated dynamically to programs as required, and then released. The memory was Core Store, except on the later S & T series machines where electronic MOS memory was fitted. The top of range 1906A and 1906S machines normally had a Paged Memory Unit fitted and programs had access to 'virtual' memory. This was available as an option on the 1903T, 1904A and 1904S, although I never came across a system so fitted; in my experience GEORGE 4 was fairly rare. The 1904E, 1904F, 1905E, 1905F along with the 1906E/F and 1907E/F (the dual CPU versions of the 1904E/F machines) had a CPU µcode in a tray underneath the CPU logic similar to core store, but read-only, having ferrite toroids only where a binary '1' was needed. The FP6000 and 1904/5/9 used a µProgram but as dedicated DTL logic. Some texts describe the technique as 'one-hot' where a binary (bistable) was set for each particular 'state' of the instruction. - This was much faster that the E/F series core, even on the 1904/5/9. The fixed core store on the 1904E series was taken from the 1906/7 machines that used the same technique. The later West Gorton CPUs from the 1904A onwards to the 1904S also used the 1904 'one-hot' µProgram as it was faster. - The 1906A/S used ECL logic gates and an asynchronous design, allied with a pipelined architecture that allowed up to four instructions to be being serviced concurrently. A 48-bit data highway to/from store and the use of store interleaving also boosted store accesses down to as little as 240nS for the 1906S and its Plessey plated wire store. All of the Stevenage systems 1901 through 1903S also used the 'one-hot' approach, allowing a better compromise between cost and performance. There was no provision for altering the microcode, unlike on the 'New Range' 2900 series machines. In some cases different models were a common processor with a different timing board. For example the 1903T/1904A(Mk.2)/1904S were the same processor with a different timing board. All machines, except for some 1901 models (hand switches and lights only) were controlled via a console typewriter, which was a modified KSR33. The speed rating shown is an approximate relative speed of instruction execution, with a 1906S being taken as 1.00, and is roughly equivalent to an IBM 370/155. The 1901A (0.02) is roughly equivalent to an IBM 360/30. Instruction timings for some models can be found here.
CONSOLE TYPEWRITER
|
||||||||
