Then There Were Three
Of the remaining two RISC architectures left to share the high end 64-bit MPU arena with IA64, the one most threatened by Alpha’s exit is Sun Microsystem’s SPARC. Sun is currently in what should be the strongest phase of its product cycle with the ongoing deployment of its new UltraSPARC-III (US-III) core throughout its product line. Unfortunately for Sun, their new processor design is unusually under powered even by Sun’s unenviable track record. The next entry on the SPARC road map is the US-IV, which is a process shrink of the same basic US-III microarchitecture. It remains to be seen how well this deeply pipelined, in-order processor will scale in performance with higher clock rates. The next big thing for Sun is the UltraSPARC-V. This all-new design is rumored to be a high frequency out-of-order execution superscalar processor. Some reports even suggest that it will incorporate some form of hardware multi-threading, possible even simultaneous multi-threading (SMT).
There is no reason to believe that the US-V couldn’t be competitive with IA64 given similar implementation quality and semiconductor process technology. However history has shown that out-of-order execution is a technology that is difficult to master and the first out-of-order machines in the MIPS, PA-RISC, and Alpha families experienced lengthy schedule slips and protracted and painful rollouts. Combining a move to both out-of-order execution and SMT in the same processor generation would be a daring gamble given Sun’s multi-year schedule slip for the prosaic US-III design. Even the incomparable computer design explorer/pioneer Seymour Cray made a point of trying to avoid taking on more than one major technical risk per product generation.
So it seems that Sun’s success or failure competing with IA64 will, to a large extent, rest on the shoulders of the US-III and its derivatives. Given the Alpha technology that Intel has just acquired it seems that relying on the ability to build large scale computing systems using SPARC may be a relatively short lived competitive advantage. Sun’s historical strength has been in its software base. If Intel’s 64-bit architecture starts to show signs of eating into the SPARC lead in application software, then Sun will become vulnerable to growing economic pressure to adopt IA64 and cease spending money on future SPARC products.
Up until the announcement of the termination of Alpha, Compaq and IBM were following the same twin track approach of embracing both IA64 and their own 64-bit RISC architecture. The road maps for both proprietary processor lines showed the same plan to differentiate themselves from IA64 by diverging from historical trends in processor design, and instead aiming straight up to the stratosphere in terms of processor bandwidth and system level scalability. This common strategy was embodied in the POWER4, EV7, and EV8 processors. That is not to say that these devices would be laggards in computing power. All three would likely have more than matched contemporary IA64 devices in processor performance as well as totally eclipsing any realistic estimate for future SPARC devices.
Although Compaq’s failure to stay the course may give pause to IBM management, it is unlikely to influence their plans. Unlike Compaq, IBM has a rich tradition of designing, manufacturing and deploying its own processors. This capability is supported by a large and competent research infrastructure capable of achieving both scientific advances in basic solid state physics, and exploiting this knowledge to push the state of the art in new and existing semiconductor technologies. In addition, IBM know-how in the area of physical packaging of semiconductors for large-scale systems is unmatched. IBM’s greatest technical weakness is design conservatism. It relies extensively on highly automated, synthesis-based design methodologies in its MPUs  . A recent paper indicates that IBM engineers have woken up to the exciting possibilities afforded by aggressive innovation in circuit level design . But they are years behind Intel and Alpha designers in practical experience.
The POWER4 will likely come to market late this year and dominate all comers in most system and processor level performance benchmarks until the Alpha EV7 ships. Even the newly strengthened Intel will take years to field IA64 processors capable of challenging POWER4 and its descendants across the complete spectrum of commercial and technical applications. The real challenge for IBM will be to achieve and retain a large enough customer base for POWER4 to justify the ongoing expense of keeping it competitive in face of Intel’s potentially decisive economies of scale for IA64. One possible strategy IBM might explore is to try to convince Apple to give up its dual sourcing policy for processors in return for delivering the POWER4 microarchitecture in a device and package form factor suitable for Macintosh desktop computers.
Discuss (16 comments)