Intel on AMD's early dual-core wins: "Not so fast"

Updated: Literally and figuratively.

Still smarting from having to swallow its pride over the success of AMD's 32/64 hybrid technology (AMD64), Intel appears once again to be on the short end of AMD's technological stick -- this time, over dual-core chip technology (the technology that basically packs two CPUs into one chip). 

Unlike with the hybrid technology, Intel was first to market with a dual-core chip -- the Pentium Processor Extreme Edition.  In bloodline terms, the "EE" is really a Pentium 4 -- a 3.2 GHz part that's based on Intel's Smithfield core, includes Intel's hyperthreading (HT)  technology, and that's been coupled with Intel's 945 chipset.  Though not as advanced in features as the chipsets that Intel is coupling with its next wave of dual core desktop chips (the 2.8, 3.0, and 3.2 GHz Pentium Ds, due this month according to recent statements made by Intel president Paul Otellini), the EE/945 combo allowed Intel to ship a dual-core chip that served the needs of certain enthusiast segments -- particularly gamers -- where the software designers were more likely to take advantage of the HT technology. (HT support must be explicitly supported by the developer in order for users to take full advantage of it.)  Among the manufacturers to jump on the April 18, 2005 dual core release with systems targeting gamers were Dell, Alienware, and Falcon.

But, while AMD may be finishing in second place in terms of timing, the company is scoring points on a number of other fronts with its dual core technology, otherwise known as X2.  According to CNET's first wave of reviews (as well as those of ExtremeTech and Toms Hardware), AMD's Athlon 64 X2 has knocked the rest of the desktops out of the ballpark.  The early results make it appear as though AMD could once again be trumping Intel.  Not only did the AMD X2 outperform the Intel EE on most tests (a testimony to AMD's design considering that the X2 was working with slower memory), but current Athlon 64 system owners will be overjoyed to know that they don't have to go out and buy a new system or motherboard to get the benefits of the X2's performance.  Instead, the do-it-yourself set need only pop out the old chip and replace it with the new.  This cut bait and run approach simply isn't available in the Intel world.

But, as the headline to this blog suggests, Intel is hinting that we shouldn’t be too fast to pass judgment and that time may tell a different story.  For starters, according to Intel, AMD's integration of the memory controller with the processor via its Direct Connect Architecture (and the way that results in superior performance) may appear to have been a prescient design choice now that AMD is winning benchmarks with components that look like tortoises (eg: DDR memory) compared to Intel's hares (eg: DDR2 memory).  But sooner or later, Intel warns, AMD's design choice will run out of gas and AMD will be forced to make changes on the chipset and memory controller fronts in order to keep up with processor enhancements.  When it does, says Intel, buyers and OEMs of AMD product will have to requalify the entire platform while Intel will already be in market, for example, with DDR2-based systems (ones that it thinks will seize the performance lead) and cranking on the volume.   Indeed, a story in today's Digitimes indicates that a complete overhaul of the AMD platform is on its way. 

Brute force may be another reason that Intel is coming up short.  Today's software lacks the necessary multithreading finesse to take advantage of technologies like Intel's HT.  So, in general, benchmarks will reflect the sheer brute force capabilities -- the integrated memory controller for example -- of the design.  Intel's Pentium EE may phsically be a dual core chip, but logically, becasue of the HT technology, it's really four cores.   For now, though, the software isn't good enough to make the EE behave like four cores.  But, over time, as evidenced by inclusion of similar threading technologies in competing chip architectures such as Power and SPARC, that lack of finesse is expected to be overcome by the prevalence of more sophisticated software development tools and techniques.  Once those tools and techniques become more prevalent, Intel is expecting that it will be the equivalent of Star Wars' Han Solo kicking the Millennium Falcon into hyperspace. (And it will probably happen with the requisite false starts too.)   One reason, says Intel, that it will be like hyperspace is because the aforementioned factor of two that turns a two core chip into a four core chip (logically) will eventually be a factor of four, then eight, and so on. 

To compete, it's safe to assume that AMD will introduce something that's HT-compatible, but so far it's not in the current gear.  Nor is it in Intel's Pentium Ds.  For all intents and purposes, the EE and the D are the same chip built on the same core (the Smithfield core), but the big difference will be in the chipset.  Here again, Intel is introducing a bevy of built-in features such as support for high-definition audio, RAID (redundant array of inexpensive disks), PCI Express, and Vanderpool-like virtualization. (AMD has its own virtualization support coming in a technology called Pacifica.)  Intel claims that with these features, the "D" platform will be targeting the mainstream users (as opposed to the way the already released EE targets the enthusiast).

Finally, Intel is arguing that it has so many dual core projects in the works (15 in all) and that dual core will be so integral to everything Intel does (from mobile to desktop to servers), and that its fab technologies will be so geared towards volume (eventually extracting 65 nm designs from 300 mm wafers) that the chipmaker will be able to serve the mass market in ways that its competitors cannot.

So, will Intel have the last laugh?  The last time I said that it would (in the pre-Opteron days), the AMD fanatics in ZDNet's readership took me to the woodshed for a well-deserved spanking.  AMD clearly got Intel's attention after Intel finally responded with a 32/64 bit hybrid design of its own.  So, having learned my lesson, I'm not one to so easily count AMD out again.  Not only that, AMD's director of server and workstation marketing Pat Patla clearly sees some of Intel's positioning as FUD (fear, uncertainty, and doubt).   

Regarding the volume issue, Patla told me via email "Sales of AMD Dual-Core [Editor's Note: "Dual Core" stricken from AMD's original statement at AMD's request] Opteron processors, introduced just three weeks ago, are already exceeding expectations, yet we foresee no capacity limitations.  Manufacturing capacity is simply not a concern for AMD."  On the memory design issues, Patla was equally bullish about DDR's future, and reiterated the advantages of processor-swapping, saying "DDR memory will be available for years to come.  By upgrading to AMD Dual-Core Opteron processors, DIY users can see 30- to 90-percent performance increases without having to trade-in their platform."  And on the issue of whether Intel's HT technology will eventually give software the equivalent of hyperspace, Patla says "HyperThreading, like large caches, is needed by our competitor to account for architecture inefficiencies.  Most major server vendors turn HT off when they run benchmarks, and independent on-line reviewers have shown that the benefit of HT in Intel's dual-core P4 EE product is almost negligible. AMD Dual-Core Opteron processors exceed the performance increases HT only promises."  This last statement is currently (emphasis on currently) true and pretty much jives with what Intel is saying.  What isn't clear is what the future holds.  It never is.