P4 to 10 GHz in 3 Year Timeframe

Intel continues to beat the frequency drum. The company is extremely confident on the yield of its high frequency parts, and demonstrated a 3.5 GHz P4 at IDF. Sources have indicated that Intel plans to be releasing a 4 GHz Pentium 4 by the third quarter of 2002 (internal ALU operation is already at 4 GHz on the 2 GHz Pentium 4). We'll remain a little skeptical because a lot depends on market conditions, and uptake of existing SKUs, as well as Intel's fabrication resource planning. Market conditions often impact how and when Intel will roll out a new product, as much as technical concerns. A slow PC economy may leave Intel thinking conservatively about giving up fab capacity to new products. What is clear is that Intel is accelerating its own roadmap to put the Pentium 4 out of reach of any competitors. Intel pointed out that the transition from a 1 GHz Pentium 4 to a 2 GHz part took about 4 months. This is a much steeper climb than in previous P6 and P5 architectures where frequency jumps of such significance took years to achieve. Intel also acknowledges that the number of effective instructions per clock is not increasing at the same rate, the rate of increase having been higher on previous architectures. If we assume that performance is a product of frequency multiplied by instructions per clock, it is clear that higher frequencies are necessary to maintain the total performance throughput.

This is why Intel is also putting so much emphasis on the work it is doing inside the chip to handle prefetch prediction, and to take advantage of the micro-op execute pipeline of the Pentium 4. The deep execution pipeline of the Pentium 4 demands that typical IA-32 Assembler be converted to micro-op code that takes advantage of the greater number of physical registers inside the Pentium 4 however, this also leaves the chip vulnerable to mispredictions that flush out the pipeline, and impact the performance advantages of the Pentium 4 architecture. The true capabilities of the Pentium 4 will no doubt be realized as a result of Intel's evangelism, and extensive research into optimization tools and compilers, but Intel will need to keep the GHz pressure up to ensure it maintains headroom on its chips.