In June 2002 at Computex Taipei trade-show AMD claimed that there were no problems with demonstrated x86-64 processors and their fabrication process despite of the fact that the CPUs ran at 800MHz, far below 1.60GHz and above, the target frequency of the first commercial Hammer chips. The company continued to state there were no problems at all in October, when it needed to explain why the processors are postponed again. Some of our sources in the company said there were no problems with yields of x86-64 processors, except the problems with achieving high enough frequencies even earlier this year. Well, today I learnt from a BusinessWeek’s article that AMD paid IBM $46 million in November for solving the problem with SOI process technology.

Analysts expected possible problems with silicon-on-insulator at the time when AMD announced intention to use this processor, however, the company only indicated confidence and tried to convince everyone that there would be no problems at all. It is now clear that the company underestimated the possible challenge and paid $46 million for it (it was indicated in the fourth quarter report). However, the actual problem is not in paying large sums, but in the fact that AMD only paid when the situation was critical for them. It is now clear that there were problems with the chip in June 2002 and even earlier. To tell you the truth, I doubt that AMD engineers, who are supposed to be really experienced, did not figure out the source of the problem in Spring 2002; if they did, executives should have done anything, but to solve the issues and begin the mass production of x86-64 processors as soon as possible. Why didn’t they do it? By now AMD has not only lost $46 million, but a lot more on Hammer delay. I think that is a big question and only AMD executives can answer it.

As we see from this example, the main idea of running a semiconductor company these days is to make the right decisions, not to try to prove your technical leadership or develop very advanced products. Another couple of examples of such idea is NVIDIA’s NV3x family: while ATI has just announced the second generation of its DirectX 9.0 products, NVIDIA cannot bring its first gen to the market, although, it is said that NV3x is more feature-rich compared to R3xx; Intel tried to surpass all chipset maker by rolling-out technically innovative i820 core-logic, but was outdone by VIA. I can recall other examples, when companies tried to outdistance the rivals, but failed, though I think you got the idea.