CES 2009: With Dirk Meyer and Phenom II, AMD keeps on trying
Will AMD's change in leadership and Phenom II allow it to better compete with Intel as economic crisis continues to grip the globe and PCs increasingly become commodities?
By Brian Dipert, Senior Technical Editor -- EDN, January 12, 2009
AMD President and CEO Dirk Meyer's keynote last Thursday may not have been well attended, but it was well worth my attendance time if only for the pragmatism it provided. The poor attendance was in part, I suspect, a reflection of poor CES logistical planning; AMD was at the Hilton and started at 11am, whereas the tandem keynote from Consumer Electronics Association President and CEO Gary Shapiro and Sony Chairman and CEO Sir Howard Stringer was halfway across town at the Venetian and didn't wrap up until past 10am. For that reason, I skipped the Shapiro-and-Stringer show to ensure I could catch Meyer's gig, and judging from post-keynote feedback I made the right choice.
Meyer began with perhaps the crispest definition of the company's long (and often confusingly) touted Fusion strategy that I'd heard to date. He pointed out that there are only two companies in the world that can deliver world-class microprocessor technology (at least in his opinion ... IBM might disagree with his AMD-and-Intel-only focus), and only two companies that can deliver world-class graphics technology (ie AMD via its ATI Technologies acquisition, and Nvidia. We'll have to wait and see if Intel's Larrabee gets added to the list in the future.). In doing so, he was echoing AMD's "total solution" advantage that I'd pointed out in a writeup published earlier that morning. Early on in its Fusion promotion, AMD mistakenly over-fixated (or, depending on your perspective, allowed the media to over-fixate) on the single-die "Holy Grail" integration of CPU and GPU technology afforded by the ATI acquisition. As the schedule for the single-die implementation repeatedly pushed out, the predominantly non-technical press translated the delay into a perceived failure on AMD's part, when in reality AMD's all-under-one-roof Fusion advantage remains conceptually valid regardless of whether the implementation is multi-package, multi-die in a single package, or the inevitable eventual single-die.
Meyer also noted early on (again, echoing themes I'd explored just a few hours earlier) that while he could give a traditional Moore's Law stump speech, he wasn't going to do so. "No question, this is a challenging time for all of us across the supply chain," he noted, in acknowledgement of the economic crisis gripping the globe. However, "Regardless of how the economy fares ... we are in the midst of a sea change," regarding the implementation and adoption of technology. PCs are increasingly becoming commodities, and their CPUs have largely hit a thermal-induced clock speed ceiling. Throwing more and more transistors at the performance problem in the form of increasing core counts, burgeoning cache sizes and the like is producing diminishing returns. In general, I found Meyer's candor refreshing, and a welcome contrast to the muddled style of his predecessor, Hector Ruiz.
With that said, to conclude that increasing core count, burgeoning cache sizes, and the like are no longer relevant would be an over-interpretation of the current situation. Witness, for example, the Phenom II microprocessor AMD unveiled at the show. Back in September of 2007, AMD unveiled the big, buggy quad-core Opteron (aka Barcelona) CPU for servers and workstations, built on 65-nm process technology, and the company followed up two months later with the equally underwhelming Phenom consumer processor. It's not surprising, therefore, that the Phenom II (and bigger-picture Dragon platform) consumer spin of the 45-nm Shanghai server CPU showed up two months after the latter's release. Below I've provided a list of some of the Phenom II reviews for your perusal:
-
AMD Phenom II X4 940 & 920: A True Return to Competition (AnandTech)
-
AMD Phenom II X4 940 Black Edition Review (ExtremeTech)
-
Enter The Dragon: AMD Phenom II X4 940 (HotHardware)
-
AMD's Phenom II 940 guns for the Core 2 Quad Q9400 (The Inquirer)
-
AMD Strikes Back with Phenom II -- Full Analysis and First Benchmarks! (Maximum PC)
I thought Slashdot's headline summarized the situation well: "45-nm Phenom II Matches Core 2 Quad, Trails Core i7." Said another way, while Phenom II is performance-competitive with equivalent-priced Intel Penryn CPUs fabricated on 45-nm process technology and based on the company's prior-generation Core microarchitecture, more recent Core i7 devices derived from the Nehalem microarchitecture still run rings around it. You might therefore think I'd be concerned about AMD's chances, especially considering the high-level architectural commonality between the competitors' latest-generation products, including a large pool of core-shared L3 cache, a dedicated core-to-core interconnect link, and on-die DRAM controllers.
You might think that, but I'm actually not terribly concerned about AMD's competitive fortunes, at least for the next quarter or few. Fact is, Core i7's performance improvements accompany a notable price up-tick as compared to both Penryn siblings and Phenom II alternatives. As Meyer noted in his keynote, the economic malaise is compelling customers to consider economic factors as a primary (arguably, the dominant) purchase criteria. Therefore, Intel's faced with an ironic product line mix. Whereas it's well positioned in the entry-level netbook and notebook segments with Atom, it's (temporarily, at least) overshot the high-end market demand with Nehalem. Abundant news reports suggest, in fact, that the company's customers are exhorting it to slow down the rollout of both new desktop and notebook PC-tailored products, because they still have abundant unsold inventory of current-generation systems.
One aspect of the Phenom II reviews I perused particularly intrigued me; the substantial degree of overclock capability reported for clock-unlocked Black Edition versions of the chip. I had lengthy briefings with AMD's graphics, notebook, and desktop PC groups after Meyer's keynote, and in the latter meeting I inquired about the overclock situation. If the chip was truly capable of notably higher speeds than its 3-GHz specification would otherwise suggest, why wasn't AMD productizing faster speed bins? The pragmatic response I got suggested that because AMD had been over-aggressive with Phenom and consequently got burned by both process and product unpredictability, the corporate psyche pendulum had perhaps now swung too far in the over-conservative direction.
AMD's spokespersons were confident that the company now had a better handle on the Phenom design capabilities and limitations in this second-generation litho shrink of the microarchitecture, that such process and product impermanence would hopefully not make a repeat appearance this time around, and that the company would therefore regain its former confident spec swagger as it regained cultural confidence. I also heard that AM3 socket-supportive DDR3 SDRAM-capable versions of Phenom II would arrive "soon," as would triple-core (ie one core disabled) variants, but that the decision of whether to support dual-core demand via a dedicated design or a two-core-disabled SKU (or both, at different points in time) had not yet been made.
Near-term situation aside, and customer resistance to the contrary, the long-running tech evolution treadmill will compel Intel to proliferate the Nehalem microarchitecture into lower-priced CPU segments in the coming months and quarters. As such, if Phenom II hasn't notably ratcheted up the performance notch a year from now to match its competitor's price/performance evolution, I'll probably be singing a much more pessimistic tune. But for now, I'm cautiously optimistic that while Phenom II probably won't enable the company to gain much, if any, ground on Intel, AMD should still be able to hold ground and stay in the game for another inning. See, AMD fanboys, I'm capable of being even-handed after all.
