[quote]Originally posted by bauman:
More realistically, though, what about this: Ignoring Altivec, other stuff like that and the rest of the computer, a vast generalization could be made that MHz and Pipeline length are the only two determining factors in processor performance.<hr></blockquote>
No it can't.
Pipeline length contributes towards latency.
Consider an idealized processor; it has an input (where it gets instructions and data) and it has an output (where it emits answers).
It consumes one unit of input per cycle, and excretes one unit of output per cycle. That is, it completes on operation per clock cycle.
Now pipeline the processor.
It still completes on operation per clock cycle. It adds latency -- a clock cycle for each pipeline stage (with some provisos).
But it doesn't alter the raw number of instructions completed per cycle.
As long as it's spitting out one finished instruction per cycle, the pipeline length is immaterial.
[quote]Since each stage in a pipeline takes one clock cycle to complete, you could say something like this:<hr></blockquote>
Yeah, you could.
You'd have to be an apologist of Everyesque proportions to do so, and it would have no correlation to the real world, but I suppose, yeah, you could say something like this.
[quote]G4 = 7 stages ÷ 800MHz = 7 ÷ 800,000,000 Hz = 0.00000000875 s = 8.75 x 10 ^ (-9) seconds per instruction = 8.75 seconds per billion instructions. Or, alternatively, 0.114 GigaFlops<hr></blockquote>
Except on the right code mix the thing can retire >= 1 instruction per clock cycle. Just like every other processor made in the past, what, seven years?
[quote]P4 = 20 stages ÷ 2,000,000,000 Hz (2GHz) = 1 x 10 ^ (-8) seconds per instruction = 10 seconds per billion instructions. Or, 0.100 GigaFlops<hr></blockquote>
Or not, because it too can retire an instruction every clock cycle.
[quote]Obviously, these figures are not accurate in the real world, but I wonder if they correlate to each other accurately, so, <hr></blockquote>
Well, no, they don't. Because they have no basis in reality.
[quote]Wow. Isn't that funny. I got 2.86. I didn't even think that I would get close to 2.8, but I thought It would be interesting to approach it from a different angle.<hr></blockquote>
And 2.8 is magic... why? Doesn't take 2.8 GHz of P4 to cream a G4 at everything except for goddamn RC5.