uhler PDP-10 Product Options 19830407

This interoffice memorandum, dated April 7, 1983, from Mike Uhler, addresses questions regarding the Jupiter II project's decision to build a CPU with four times the performance of a KL10, rather than a less powerful, quicker-to-market alternative.

The memo evaluates two main options: a KL10 conversion or a variant of the Jupiter architectural design. It argues strongly against a KL10 conversion due to several factors:

• The current team lacks experience with the KL10 design.
• The KL10 design has numerous bugs (especially in extended addressing) that would require significant redesign efforts, making it more than a simple conversion.
• The KL10 has physical memory addressing limitations (4 Mwords) and suboptimal I/O interfaces.
• Converting KL10 to newer technology (10KH or MCA) would only yield a 1.5-2.0x performance increase.
• A KL10 conversion to gate arrays would be complex, requiring extensive verification and adding an estimated 9 months to the schedule.

In contrast, the Jupiter architectural design is presented as superior, having benefited from an eight-year learning curve and solving many existing KL10 architectural problems. Its enhancements include:

• Full 30-bit virtual address space support.
• Correct implementation of the full architecture.
• Improved hardware monitor interface and timesharing efficiency.
• Increased functionality and designed-in interfaces to corporate buses.

Regarding the performance goal for a Jupiter design, the memo suggests that a 2-3x KL10 machine is relatively easy, a 4x KL10 is "within reach" with careful design, and a 5x KL10 is "quite difficult." Crucially, it estimates that a 4x KL10 Jupiter design would require no more than 6-9 months additional schedule compared to even a simple (but problematic) KL10 conversion.

The memo concludes that pursuing a PDP-10 design based on the Jupiter architectural design with a nominal performance of 4x a KL10 represents the optimal tradeoff between performance and time-to-market, providing a superior machine without a significantly longer schedule than a problematic KL10 conversion.