11.2.07

AMD’s Apotheosis

In mid-2003 AMD did what was up until then unthinkable in the commodity microprocessor market: it challenged the monopoly CPU maker Intel by providing a starkly different CPU roadmap while holding only a fraction of CPU market. Intel up to that point tried to transition the market from a 32bit to 64 bit computing by introducing a new architecture, dubbed Itanium. Alas, this new brand spanking new CPU architecture was incompatible with the old x86 instruction set and would essentially require a rewrite of all software to run optimally. Itanium’s high price tag, lackluster overall performance and Intel’s arrogant assumption that customers would blindly follow their technology roadmap eventually opened up a window for AMD.

AMD responded to customer’s thirst for the next generation chip by introducing the Opteron processor that included 64 bit extensions – the AMD64 instruction set. Among other things, the processor was fully backward compatible with the old x86 instruction set. However, the chip was architected with enterprise class features like an onboard memory controller and hyper transport bus that allowed high-speed multiple CPU interconnects. With a team that included many from the vaunted Alpha chip team, the Opteron took AMD to the next level. By 2006, AMD owned a quarter of the CPU market and an even higher fraction of the server microprocessor sub-market. In benchmarks, Opteron ascended to the throne and was king for over 3 years. Eventually, even the most technically ignorant manager could not justify the purchase of Intel’s dead-end products.

Intel’s strategy was clumsy precisely because it bifurcated the processor market into two camps: consumer CPUs would utilize the x86 instruction set while the enterprise sector was to adopt the Itanium instruction set IA64. To this day, Intel stubbornly clings onto the Itanium chip – although it is de facto a dead architecture. However, someone at Intel finally woke up and realized that chasing gigahertz via the discredited NetBurst architecture would end in failure as heat issues and diminishing manufacturing yields were proving. Just in time came the Core architecture and a slew of new chips from the Israeli design center- normally designated to design chips slated for the lowly mobile market.

Essentially Pentium Pro on steroids, the Core microarchitecture is slated to replace the discredited NetBurst architecture throughout Intel’s product line. The Core 2 Duo has managed to unequivocally help regain the benchmark lead in the consumer sphere and the Woodcrest in the two-way server market. To win over the high-end enterprise market, Intel has to jettison the front-side bus with something akin to the common systems interface (CSI) present in the Itanium. Rumored to be introduced by 2008, the CSI may finally turn the Xeon into all-benchmark champion and bona fide enterprise level chip.

AMD has to respond to the awakened giant by mid-2007 by answering the fundamental benchmark challenge and the ever increasing multi-core requirements. Already behind by six months with its quad core processor, AMD is also behind in manufacturing capabilities (it still hasn’t transitioned to a 65nm process from the current 90nm). While AMD’s improved K8L architecture, to be introduced this summer, will provide some relief with true quad-core processors, the real battle is yet to come…