Barrett still has some fight in him

(continued from previous page)

Photo: Gartner

Intel CEO Craig Barrett apologizes for the company's failure to deliver a 4GHz Pentium processor.

[PVRK]

There is a limit to everything

It is obvious during the last few years that just increasing the clock speed is not going give any performance boost. But Intel deliberately masked that with their propaganda while AMD and other concentrated on increasing performance while ignoring the GHZ hype. When Pentium 4 was released many people saw the Intel GHZ game as it came with much longer pipe than necessary. I am glad that Intel finally stands exposed and consumer PCS will not get obsolete every 4 months. But I always wonder why press did not expose their GHZ myth.

[Tex Murphy PI]

Hind-sight is 20/20

While the march towards higher clock-speeds (for marketing reasons) was the principal drive for the Pentium IV architecture, it is only one of the reasons why Intel went down that path.

Technically, it is far easier to speed up the processing sream in a serial, versus parallel manner (easier get two cars speeding down two lanes at the same speed, compared to ten). Much of today's architectures are going down that path - look at SATA (serial ATA) as an excellent example.

Making a chip run faster is fruitless unless the supporting architecture supports it. This is why Intel pushed for Rambus (also a serialized memory architecture) and SATA to be adopted.

If you studied the Pentium IV (especially the Prescott) architecture, you will notice that efficiency actually improved with greater frequency speed bumps. To the point where it trounced any Northwood chips above 3.6GHz.

Depending on the type of applications you ran, the Prescott beat out the newer Athlons, but also lost to it in other applications. Notice how the P4 does great on programs that "stream", while the Athlons excell at parallel tasks? Different architectures excell at different tasks.

Intel's mistake is two-fold.
1) They assumed that multimedia "whiz-bang" stuff sold processors. Wrong. Primary computer usage is for data processing and logic. Most real world applications do not stream well - like video - and choosing to make the P4 the best in multimedia (instead of data processing)was a VERY BAD CHOICE.
2) They assumed that brute force will be able to overcome the loss of clock-cycle efficiency due to streaming. This would have worked IF they were able to circumvent the horrific power requirements AND heat dissipation.

Had Intel been able to conquer the age-old problem of heat dissipation and power consumption, nobody would have cared about the lower instructions per clock cycle disadvantage to AMD's Athlon. However, to do that, Intel had to release chips that were about twice as fast as the AMD equivalents, and that would have been impossible - since they would have to bump up their speeds at twice the rate AMD did to keep even.

Ultimately, Intel's problems lay within their bad judgement of what technology tree to take. Faster Serial, or Slower Parallel. They were correct in assuming that the best way to overcome some of the subsystem limitations was to serialize it. But they were incorrect in putting that into the CPU architecture.

Pvrk Prasad mentioned that the exposure of the "Intel Propaganda" and the cessation of the GHz race will stop PCs from becoming obsolete every 4 months. Sorry, sir! That will ALWAYS happen, regardless of the GHz race.

It's such a pity that Intel's leadership was either too blind, over-optimistic or just plain ignorant to have seen it until today.