[citation][nom]freiheitner[/nom]@thearm, actually the P4 was pretty bad for its time. Benchmarks on Tom's Hardware at the time showed a (lower clocked) Pentium III beating the (higher clocked) P4 on I believe 5 out of 8 tests, and the Athlon beat the P4 on 6 out of 8 tests. Was the P4 good enough to run Windows and Office? Sure, but so was the Pentium II based Celeron. P4 required higher clock rates to get the same performance as other chips, and I believe that was TA152H's point (though I don't mean to put words in their mouth).[/citation]
That's kind of what I was saying. I agree with you, and disagree with him, that the Pentium 4 was a very good chip for its time. In my opinion, it was horrible compared to other chips. If we compare it to the K8, it was much larger, much slower, and consumed much more power, despite Intel's superior manufacturing technology. I don't see how this is good, and apparently Intel didn't either since they killed it.
I also agree with you that it could run most apps fine - I still use a Tualatin overclocked to 1.6 GHz most of the time, and it works fine. But, even doing that, the Pentium 4 gets way too hot, and uses way too much power. There really isn't one metric where it can be called superior, or even mediocre, except for clock speed, which equated to sub-standard performance anyway.
By contrast, Nehalem is broadly excellent at almost everything. So, it's a much harder act to improve on. Still, I have high hopes for Sandy Bridge, but I think the expectations Otellini is creating are just unrealistic and probably will lead to many people being disappointed.
The 286 improvement will be much higher. The 386 sucked. The 486 was also a huge improvement. Pentium, maybe a little less, but still quite large. It has no hope of matching any of those three. The Pentium Pro comparison is ambiguous, since it actually ran the most common code from that time slightly slower than the Pentium (meaning 16-bit), which was somewhat fixed in the Pentium II, and completely fixed by software moving to 32-bit (although it was not really a 16-bit or 32-bit issue, just segment registers were used extensively in Real and 286 Protected, and not generally used in 386 Protected mode).