Penryn Goes Public: Competitive And Other Closing Thoughts

Continued from ‘Penryn Goes Public: More ASPs‘…

Perusing my peers’ benchmarking results, the anticipated Penryn performance boost over Core 2 generally pans out; keep in mind, too, that these are introductory 45 nm clock speeds. To wit, I’ll show you a photo, also taken at IDF of another Intel ‘Skulltrail’ demo platform, this one vapour cooled and otherwise tweaked for maximum speed:

Cool case, huh? Ironically, as the Inquirer points out, since the system is based on dual Nvidia Media Communications Processors (akin to the architecture of the AMD Quad FX aka 4×4 system I used earlier this year), the motherboard sports an AMD-developed HyperTransport link. But, to the point of this particular writeup, the CPUs are now reportedly running stable at 5 GHz (they were at sub-4 Ghz at IDF).

Where’s this all leave AMD? Reportedly, the company will launch Phenom, the consumer-targeted desktop spin of the single-die quad- (and tri-) core design that previously birthed the ‘Barcelona’ workstation and server CPU, next Monday (I can’t yet publicly confirm or deny any of this at the moment, of course). Phenom is reputedly going to crack through the underwhelming 2 GHz clock speed barrier that the quad-core Opteron launched at (and seems to still be having trouble delivering at), but only barely…and at higher TDPs than originally forecasted.

Intel’s one process generation ahead of AMD, with requisite performance, power consumption and die size/cost benefits incurred as a result. From where I’m sitting, Intel’s 45 nm process also seems to be ramping much more robustly (as measured by both clock speed and yield metrics) than is AMD’s 65 nm litho. In terms of absolute performance, at least to date, I don’t see anything that’d knock Intel off the throne it wrestled away from AMD several years ago as part of the NetBurst-to-Core microarchitecture transition. The same goes for performance-per-watt, at least at the CPU level…Intel’s continued reliance on FBDIMM in the enterprise space can, depending on your workload assumption, leave it at a competitive disadvantage versus AMD and conventional DIMM-based DDR2 SDRAM.

And what about performance-per-dollar? This is fundamentally how AMD’s remained in the game the past few years; by pricing its equivalent-performance parts below Intel’s offerings. Clearly, though, this is a short-term reactive strategy, not a path to long-term fiscal stability. Will Phenom enable AMD to regain its competitive stride in the critical high-volume consumer PC segment? Keep watching this space; I’ll share my thoughts as soon as AMD’s embargo lifts.