Those questioning the long-term viability of the PowerPC® received a strong positive answer last month at the first Linley Tech Processor Conference when IBM announced the PowerPC476FP core. The 476FP, formerly known by its project name “Gallardo” (think “Lamborghini Gallardo”), will achieve 1.6 GHz in IBM’s Cu-45HP SOI hard core implementation. The 476 is available for SOC designs this month, with first production silicon in 2010. The response has been enthusiastic in the PowerPC community.
In 2005 Apple kicked one of the legs out from under PowerPC when it announced the transition of its Mac products from IBM’s PowerPC to Intel’s x86. The ripple effects are still being felt. A second blow occurred last year when Apple’s acquisition of P.A. Semi last year dead-ended the powerful PWRficient PowerPC processor. During this period the PowerPC community has also seen notable highlights. For example, Freescale’s 45nm PowerPC-based QorIQ™ P2020 Layer 2 MAC together with their MSC8156 Layer 1 PHY is a state-of-the-art solution for 4G LTE/WiMAX wireless base stations. CPU Technology has exploited both PowerPC and IBM’s unique position as the only advanced CMOS Trusted Foundry to deliver the first merchant market trusted and secure processor. Nevertheless, some performance-sensitive customers have looked longingly at Intel’s new products, particularly Atom.
At 1.6GHz, the 476FP is one of the most powerful 32-bit IP cores available. Unusually, the DMIPS rating of 2.5 DMIPS/MHz actually increased during development. Of course, DMIPS is a poor measure for a 5-issue, out-of-order micro-architecture. The processor complex also features a 1024-entry RAM-based TLB and the sixth generation PLB6 supports multi-processor hardware cache coherency.
IBM has announced a soft IP version of 476FP for late 2010 but implementation of a core of this complexity is best done by experts then leveraged in multiple projects. IBM plans to characterize 476FP across a wide frequency range – it will be important to see whether lower clock speeds can meet lower power requirements. Unfortunately, the 476FP doesn’t integrate AltiVec/VMX; the subword parallelism would have been a real benefit to accelerating 32-bit floating-point common to DoD signal processing. However, the 476FP has set the expectation for more good news from IBM’s PowerPC core roadmap.
See the entry at Linley Chips In for more information.