Woodcrest will suffer the fate of Athlon MP

Athlon MP, introduced in 2001, had an architecture similar to that of Woodcrest/Bensley. Just like Intel's 2006 design, the 2001 Athlon MP had dual independent bus, which allowed 2P operation but no more than that. Although its performance was quite good at 2P level compared to the Xeon DP at the time, Athlon MP failed to generate much interest. AMD was not considered a worthy server CPU vendor, because it didn't have a four way solution.

Today, Intel is in a very similar situation. AMD scales up to 8P 16 core, while Intel is stuck at the cheapo 2P level.

Originally, Intel planned to follow up the Sossman 32 bit CPU derived from Core Duo with Dempsey/Bensley in Q2, then Woodcrest in Q3. Intel released Sossaman32 in Q1, but it was ignored by major Intel server vendors, including HP and DELL. Intel prepared to launch Dempsey in Q2, but it has apparently been cancelled to lack of performance and performance/watt improvement. As a result, the whole Intel server line is being shattered right now, as the hot, slow and antiquated Xeon Paxville has only 50% of the Opteron's performance.

Intel's plan is then to move Woodcrest/Bensley launch ahead to June 19, 2006. Due to Intel's manufacturing inefficiencies, I can safely bet these CPUs will be delivered from Intel's development FABs. In March, Intel showed Conroe and Woodcrest samples, these were from Intel's development FABs. With Intel's so called copy-exact methodology, it would take a couple of months to duplicate the process at its production FABs. This is why we are seeing a slow Woodcrest ramp (75% by end of of 2006), despite the fact that server CPU volume is only a couple million units per quarter.

So, Intel is brushing its Sossaman32, Dempsey plans aside, and rushing to push Woodcrest out of the door, presumably out of its development FABs. I think that one of the main purposes of the rush was to stop DELL from defecting to AMD. But, Woodcrest is already too late. Intel has lost the strategic initiative even before Woodcrest is launched.

The problem is AMD has been sitting on its next generation designs for quite a while now. AMD can deliver on these designs at the time of its own choosing. Right after Intel's IDF Woodcrest demonstration, AMD decided to move its launch schedule of Rev F parts ahead. Turion X2 was launched on May 17, 2006, Socket AM2 is to be launched on May 23, 2006. AMD will demonstrate its technologies on June 1st, 2006. AMD may launch Rev F Opteron (socket 1207) on June 6, 2006.

The impending launch of Socket F (1207) is significant, as it establishes the road to true quadcore technology. As for core revisions, K8L has been taped out, AMD should be able to pull in the new core with relative ease. This gives Opteron users a comfortable upgrade path.

AMD is producing significant number of 65nm test chips right now, in an effort to tune yields at the 65nm node. Due to the limited server CPU volume, AMD should have no problem releasing 65nm server CPUs, even before the process is fully tuned for general production. We may expect to see quadcore Opterons sooner than expected.

Going forward, we can see AMD always at least one step ahead of Intel. Preliminary results show that Intel's CORE2 CPU performance will be very close to AMD K8 at the same clock speed, even according to Intel's own guerilla benchmarketing*. AMD enjoys massive advantage on areas of mutli-CPU, multi-core scalability and throughput. Just as Intel plans to catch up with AMD in 2P performance, AMD moves the bar even higher.

One of AMD's mistake was that it talked about dual core Opteron too early, which alarmed Intel into creating its dual-die solution. This time, AMD has been very tight-lipped about its next generation plans. The only hint we got was from AMD's CEO Dr. Ruiz: "It will be interesting to see the things that we're going to do later, which will again continue to force them (Intel) to react and figure out what else to do next." Dr. Ruiz also stated in a recent Jim Cramer's CNBC show that AMD will again challenge Intel to another duel. These words from Dr. Ruiz indicate the existence of some major AMD innovations that will put Intel further behind.

Compared to Opteron users, Woodcrest users will see diminshed returns from their investments. Like Sossaman and Dempsey, Woodcrest is just another stepping stone in Intel's hope to catch up with AMD. In 6 months, before enterprises can fully validate the Bensley platform, Woodcrest will become totally obsolete again, as Intel tries to copy more of AMD64's Direct Connect Architecture.

Woodcrest, like Sossaman, will have the shelf life of a banana.

*Even with Intel installing faster hard drives in its servers, the Woodcrest 3GHZ is ony 37% faster than 2.4GHZ Opteron in Intel's SunGard ACR demo. This translates to 1.37/3 * 2.4 -1 = 9.6% clock/clock performance improvement over K8 in Intel's favourite benchmark. Sun Microsystem's John Fowler commented on Woodcrest that "Five percent (performance improvement) is not particularly easy to explain."