P3 vs. P4 Clock cycle to clock cycle

[barryng]

Although the P4 series of cpu's have higher clock speeds, I understand that they are not as efficient per cycle as the P3s. The speed advantage of the P4 is achieved because because they are capable of much higher clock speeds, not because they accomplish more each clock cycle. This then begs the question, how do you compare the two processors? How much faster than a P3 does a P4 have to be to out perform the P3?

For example, my P3 is runninng at 1008 GHz. How fast does a P4 have to be so that I gain something by upgrading to it?

 

[AndyHui]

Just taking the length of the pipeline into account is one way (probably not very accurate, since it does not take into account other architectural changes), to compare the two processors.

The Coppermine Pentium III processor has 10 or 12 stages in the pipeline (sorry....it's been a long day and I can't remember exactly). The Willamette Pentium 4 has approximately 20 stages in the pipeline. With this information, it takes a single instruction at most 12 clock cycles to complete on a P3 and 20 cycles to complete on a P4.

Taking the above information, you could make a very ham-fisted approximation say, where 20/12 = 1.667; then a 1.7GHz Pentium 4 would give the same performance as a 1.0GHz Pentium III.

This applies to a single instruction, and does not take into account caches, execution trace caches, number of inflight instructions, that sort of thing.

I suggest you have a look at this thread, and read what both pm and I have to say.

This is all completely theoretical. The best thing for you to do is have a look at the benchmarks on the main Anandtech website.

 

[andreasl]

<< This applies to a single instruction, and does not take into account caches, execution trace caches, number of inflight instructions, that sort of thing. >>



Or pipelining, or OOO Execution, or ...

And here is something else funny. Take a P3 and P4 both at 1GHz. The P3 would win easily. Now take a P3 and P4 both at 2GHz. The P3 would probably win in most cases again, but the difference would be smaller (in %) than at 1GHz. Of course since the P4 doesn't exist at 1GHz and the P3 will never exist at 2GHz, such a comparison is only academic. But it shows that there is more to a processor than performance at a certain GHz.

The best technical performance comparison IMHO between 2 processors are their maximum performance on simular wafer processes. Example: P3 1GHz vs P4 1.8GHz vs Athlon 1.4GHz etc (currently).

The best comparison for end users is probably performance/cost or simply highest performance for those with big wallets.