The flip side of the accelerating acceptance of 64 bit merchant MPUs is the continued decline of traditional high end RISC based product lines. SGI officials recently disclosed that sales of its IPF based servers have crossed over those of its MIPS based gear and now account for the majority of its server revenue. HP introduced its penultimate PA-RISC processor, the 130nm PA-8800 . It represents an attractive upgrade for HP’s existing PA-RISC customers but has little inherent appeal beyond that. HP showed less concern for the Alpha customer base it inherited from Compaq. HP advanced Alpha’s omega by killing the EV79, a 130nm shrink of the EV7. To add insult to injury, HP tried to shift blame for EV79’s demise to its foundry partner IBM Microelectronics, the same foundry which doesn’t seem to have any problems manufacturing the PA-8800. To save face (and possibly to avoid legal repercussions) HP will introduce a faster speed grade of the existing 180nm EV7 later this year and slap on a new name, EV7z. The EV7z will ship at ~1. 3 GHz, a clock frequency the EV7 could have and should have been introduced at years ago.
The most striking evidence of RISC’s accelerating decline in high end computers is the recent decision by Sun to quietly kill the development of its long delayed new generation SPARC core. The new microarchitecture represented Sun’s first foray into both out-of-order execution and simultaneous multithreading (SMT) and would have initially shipped in the UltraSPARC-V. Although Sun still plans high CPU count CMP devices based on its elderly US-II and US-III CPU cores, this decision means that SPARC has effectively joined PA-RISC, MIPS, and Alpha as once important server architectures now in extended palliative care. Sun officials try to spin this re-use and recycling of obsolete designs as at the cutting edge of throughput computing, but the poor single thread performance of these cores relative to competing CISC, RISC, and EPIC MPUs will limit them to niche applications at best. Although SPARC still has a large presence in the server market, Sun’s long term future as a hardware vendor is increasingly riding on the success of product lines based on merchant processors. In addition to limited offerings based on Xeon, Sun has publicly embraced AMD’s Opteron chip. Fujitsu also develops SPARC compatible MPUs but its SPARC64 processors are only marginally faster than Sun’s UltraSPARC series, and Fujitsu has adopted IPF for use in future high end systems.
Sun’s SPARC retreat leaves IBM carrying RISC’s banner in general purpose computing as it continues to aggressively develop new PowerPC processors for the desktop and high end servers. Later this year, IBM will roll out its POWER5 which builds on the POWER4+ by adding 2 way SMT capabilities . However, even IBM has hedged its bets by selling limited lines of Xeon, IPF, and Opteron based products, as well as sustaining its 40 year old mainframe legacy through sales of its zSeries servers. Although IBM has ostensibly replaced Motorola as Apple’s MPU vendor of choice with the recent introduction of the POWER4+ derived 64 bit PPC 970, the new MPU has not visibly slowed down the Mac’s long term market share loss to x86 based PCs. With less than 2% global market share and growing disinterest from major ISVs, it is clear that IBM cannot look to the Apple Macintosh as a strong growth market for PowerPC processors. As a result, IBM is feverishly trying to expand sales of PowerPC in high end embedded control applications, like video game consoles, to better rationalize the economics of its troubled in-house semiconductor operations. This hard work seems to be paying off as IBM is slated to provide PowerPC based computing engines for the next generation of Sony and Microsoft video game consoles, displacing MIPS and x86 based devices respectively.