On Wed, Jul 22, 2015 at 8:32 AM, emanuel stiebler <e...@e-bbes.com> wrote: > Was a wonderful idea, but the performance of the 432 wasn't nearly what was > promised.
I'm updating my dictionary to quote that as an example under the definition of "understatement". See "Performance Effects of Architectural Complexity in the Intel 432" by Robert P. Colwell. TL;DR: the inherent slowdowns imposed by the 432 architecture are relatively small compared to the slowdowns imposed by hardware implementation choices. Some of the hardware implementation choices were forced by the limited die size that was practical to fabricate in the late 1970s, but others were apparently made on the basis of inadequate or flawed analysis. Release 3.0 of the 432 architecture made several significant improvements to performance within the limits of what could be done without a complete hardware redesign, including using a preallocated chain of context objects for the activation frames of a process rather than allocating a new context object on each function call. Unfortunately it was too little too late.
