For years clock-speeds of microprocessors were considered to be the most important indicator of their performance. Back in 2000 both Advanced Micro Devices and Intel Corp. have conquered the 1GHz milestone and towards the end of the year Intel launched Pentium 4 processor with unbelievable 1.40GHz frequency. But while the company believed that thanks to peculiarity of the NetBurst micro-architecture to quickly gain clock-speeds the new chips will quickly increase performance levels, it eventually emerged that the approach is not feasible and the whole clock-speed war cease to lose sense at some point.
The main peculiarity of Intel NetBurst micro-architecture was very long pipeline: 20-stage for the code-named Willamette chip and 31-stage for the code-named Prescott processor, up considerably from 10-stage pipeline of Intel Pentium III central processing unit (CPU). On the one hand, long pipelines allow processors to run at extreme clock-speeds, however, they increase branch mis-prediction penalties, which means that software has to be developed with microprocessor design in mind. Although Intel claimed that its branch prediction unit of the Pentium 4 would reduce mis-predictions considerably compared to the Pentium III, whereas Rapid Execution Engine (which clocked arithmetic logic units at twice the core-clock frequency) would offset any performance decreases and would boost raw power of NetBurst CPUs to ultimate levels, the NetBurst faced both performance and power-related problems.
Since back in the late nineties and the early 2000s many considered clock-speed as the main measure of performance, 1.40GHz – 3.0GHz Pentium 4/Xeon were viewed as something extraordinary despite of the fact that their performance was lower than expected considering extreme frequencies. However, with the emergence of AMD Athlon 64/Opteron processors in 2003 everything changed drastically. Thanks to improved micro-architecture amid relatively short pipeline, AMD64 family of chips not only quickly left Intel NetBurst-based chips behind in terms of performance, but also forced Intel to boost clock-speeds even more aggressively than before.
It was relatively easy for Intel to push clock-speeds of Pentium 4 chips upwards since in 2004 it introduced Prescott core with 31-stage pipeline, however, performance of such processors grew much slower than power consumption. Even though initially Intel wanted to reach 10GHz clock-speed with the NetBurst micro-architecture, at around 3.80GHz (for both 90nm single-core and 65nm dual-core products) thermal design power of such processors became too high – around 130W – despite of relatively low performance.
Sometimes in 2004 it became clear that the whole clock-speed war could no longer lead to tangible performance increases. The high-end Pentium 4 processors were both hot and slow. Even though their performance could be increased by introducing models with even higher clock-speeds, their efficiency and power consumption were two obvious problems. Even though performance of AMD Athlon 64 was much higher back then, it was also clear that further speed boosts would not lead in substantial increase of actual performance. In order to continue rapid pspeed scaling both AMD and Intel chose multi-core route: properly written programs can work much faster on two slow cores than on one fast core.
In 2006 Intel introduced its Core 2 microprocessors based on highly-tweaked Pentium III/P6 micro-architecture. The Core 2 Duo chips managed to offer market leading performance at just 65W thermal design power. Next year Intel canned the whole NetBurst micro-architecture from virtually all market segments.