|
|
|
|
|
|
|
Figure 4.1
Evolution of a computer processor family. |
|
|
|
|
|
|
|
|
4.1.1 Evolution of a Computer Processor Family |
|
|
|
|
|
|
|
|
As an example of the family development process, consider the product line of the IBM 370 series as it emerged over the period 19701983 (Figure 4.1). The relationship between certain variables in a machine designcycle time, machine organization, and effective access time of the memoryand the key performance metric, MIPS, is presented in Table 4.1. The first five products range in performance from 0.25 MIPS to 4.9 MIPS, and show the variations one can achieve as the nature of the design space changes. The first two machines, which were TTL designs, had very long cycle times, strictly serial organizations (well-mappedthat is, they serially performed decode, address generation, operand fetch, and execute), and memory access times that were significant compared to the internal cycle times. These products had MIPS rates in the range of 250500 KIPS (thousand instructions per second). The next product, the 158, had a cycle time of 115 nanoseconds, had a cache so that its effective memory access time was on the order of 200 ns, and had a simple overlap design; that is, the 158 attempted to overlap the fetching of instructions with the execution of those instructions. The next two designs, the 168 and 3033, which came out of the same basic organizational definition, were pipelined ECL machines that in addition used many other techniques to enhance performance. |
|
|
|
|
|
|
|
|
IBM replaced the 168 and 3033 products with new high-end products in the early 1980s: the 3083s, which are uniprocessor products, and the 3081s, which are dyadic multiprocessor productstwo processors sharing a common main memory. In this generation, the cycle time is fixed across the |
|
|
|
|
|
