Oh dear.. a correction that needs a lot of ahmm.. correcting here.
>the main one that most articles address is the pipeline and
>amount of instructions that a thread must go through to
>finish it's computing.
The "ammount of instructions" a "thread must go through" is is absolutely equal on any chip running the same code. You can count them in the binary code if you like.
>when you need to proces more commands to finish a threads
>computing, it takes longer or has a longer pathway to go
Nop, same ammount of "commands", unless you are referring to microops, there might be a difference, but frankly, I wouldn't even know which way.
>AMD runs around a dozen or so of these commands in it's pipe
> and intel runs around 30 something. so you have a ration of
> around 3:1.
Consider this: both Ford and GM produce the exact same car (the code) on a production line. Fords line only has 100 steps (pipeline length), GM has 200. In both factories, cars move from one step to the other every 2 minutes (clock). Now tell me which one produces more cars ? They will both pump out a car every 2 minutes, so performance will be identical once you get beyond that first finished car ..
Pipeline length has no impact on throughput, it only incurs a penalty regarding the ammount of "work", say the number of people working on the poduction line, which in the case of cpu's translates into heat. More importantly, it has a huge impact on branch misprediction. In our car assembly analogy, it would be like constantly missguessing what colour and options the customer wanted, and moving the car back from the end of the line to the start. Since GMs line was 2x as long, and it still took 2 minutes to go through every step, that is where it starts to hurt, since much more work is lost every missed guess.
>Despite what AMD Fanboys will tell you, Intel IS NOT playing
> a catch up game with AMD. they are not revving up their
>clock speeds to match AMD.
Oh please, both intel and AMD constantly up their clocks *and* try to improve IPC.
>AMD has ALWAYS played cathcup behind intel to meet or beet
>the intel chips
Except at those times when the fastest CPU you could buy was an AMD 286-12 Mhz (*), an AMD 386-40 MHz, a K6-233 MHz, arguably an Athlon 500, without a doubt the 700, 1 GHz, 1.2 GHz,.. the first XPs, and now A64s and opterons.
>a good example is the update from the P3 to the P4. the P3
>capped out because of heat and they needed to redesign the
>chip.
Great example, except P3 was in no way thermally limited, especially not Tualatin on 130nm.
>Look at the Intel Pentium-M using the Dothan Technology.
>this chip could blow the P4 and any AMD out of the water
>with even LESS clock speeds (highest is around 2.13ghz) and
>less heat (27w)
Spoken like a real fanboy yourself here.
(*) granted, that chip was manufactured by AMD, but sold by intel. Imagine a chip that looks like this today:
<A HREF="http://www.cpu-collector.com/photos/AMDN80L2868C2H_FL.jpg" target="_new">http://www.cpu-collector.com/photos/AMDN80L2868C2H_FL.jpg</A>
= The views stated herein are my personal views, and not necessarily the views of my wife. =