Did Alpha Have to Die?
Since Compaq’s announcement many people, generally Alpha users, have posed the question did Alpha have to die? Compaq knew it faced a potential public relations disaster on its hands and was well prepared beforehand with glossy press kits and slick power point presentations to put forth its case. And that case essentially boiled down to the premise that further support of its widely respected RISC processor family could no longer be justified in the face of Intel’s growing move into the high end 64-bit MPU market. Accompanying this predictable and self-serving explanation for Alpha’s death sentence were soothing words to reassure Compaq’s high-end system customers that their TRU64, OpenVMS, and Linux based applications now running on Alpha hardware would have a graceful and seamless transition path to IA64 platforms.
The question of Alpha’s viability has many intertwined aspects – technical viability, marketing viability, and economic viability. The technical viability of the Alpha architecture for any reasonable time scale looking forward is hard to dispute. This is especially true as it becomes more widely understood that the early hype about EPIC being the third wave in computer architecture design after CISC and RISC was wildly overblown. As the first IA64 systems ship commercially it becomes apparent that after a dozen years and billions of dollars of research, Intel and HP have made little headway in solving the problem of VLIW’s inherent unsuitability for general purpose computation. The EPIC design philosophy, especially as realized in the mind-boggling complex IA64 ISA, is likely a liability, not an asset compared to a streamlined second generation RISC design like Alpha.
But Alpha’s architectural superiority still doesn’t guarantee technical success. Two factors that are far more important than the ISA in determining a microprocessor’s performance and cost are the semiconductor process technology used to implement it and the overall design implementation quality. The Alpha design group’s past superiority in design implementation quality has diminished as its physical and circuit design techniques and methodologies became more widely known and emulated. There is little to suggest that this competitive advantage hasn’t effectively vanished in specific examples, like the many implementation innovations Intel put into the Pentium 4 design.
Process technology is the biggest question mark related to Alpha’s technical viability. Compaq has yet to ship a true 0.18 um implementation of Alpha, even as Intel starts to ramp commercial production of microprocessors built using 0.13 um process technology. Although Compaq’s semiconductor partner IBM will not lag Intel for very long in ramping its own 0.13 um process, the limited resources Compaq allocated to Alpha hurt its ability to quickly port existing designs to next generation processes. Given similar implementation quality, Alpha’s architectural advantages over IA64 are too small to make up for a full generation lag in exploitation of the latest semiconductor technologies.
Discuss (16 comments)