FPU speed mysteries

[jj2007]

Perhaps during the test, or some portion of the test, the P4 is actually running at a lower clock speed than the Celeron. If you were timing the code in clock cycles, instead of seconds, the results would not vary significantly with the clock speed.

Will test that asap, thanks. One strange behaviour of the old notebook (P4) is that the fan starts making noise when I run timed code for more than, say, three seconds; and then the speed actually decreases drastically. Apart from that, the timings are very stable.

[raymond]

I use an older P4-1.5GHz during part of the year. I also have a newer CoreDuo rated at 1.89GHz which I use during another part of the year. That newer one runs most of my apps between 4 and 7 times faster!!! It must have something to do with the internal architecture of the processor.

[jj2007]

If you were timing the code in clock cycles, instead of seconds, the results would not vary significantly with the clock speed.

Do the Timer macros measure FPU cycles the same way as CPU cycles?

Just timed it on my office desktop, a P4 at 3.4 GHz: 400-420 ms. As Raymond also noticed, the Core Duo CPU's seem to be 4x faster. I wonder how that would affect the FPU vs SSE2 debate...?

[MichaelW]

As far as I know, for recent processors the CPU and FPU share the same clock. Now that I think about it I recall reading, during the time that Intel and AMD were actively competing for the fastest processor, that Intel had compromised the P4 FPU to the point that even with the higher clock speeds of the P4 the Athlon FPU was much faster.