Yousef, RE:simple assumption that MaxFreq ~ 600*Idsat/VC, where V=operating voltage, Idsat=600 ua/um drive current limit, C=gate capacitance. The weakness in your argument is that C is the same for AMD's process and Intel's process. They are not. Furthermore, this value C has nothing to do with the geometry of a single transistor, but is a complex function of the parallelism of the design and is also affected by factors such as layer thickness and layout. Furthermore, if you compare the Pentium II with a K6, you will come to the conclusion that Idsat is higher for the Pentium II since its current requirement is more than proportional to the ratio of the number of transistors times the ratio of the operating frequency. (The only reason a Pentium II survives is its superior thermal design. This and only this is what enables it to run at 300 MHz -- gobs and gobs of AMPS.) In short, your "constants" aren't!!!!
To prove this, lets compare the performance of the DEC Alpha processor, which runs at 533 MHz and also uses a 0.35um process and runs at 2 volts.
533 ~ 600/(2.0*C) for DEC Alpha 21064 processor
300 ~ 600/(2.5*C) for Intel Pentium II
Obviously, either the proportionality constant or C is different for both equations to be true!
According to your formula, the DEC Alpha should only be capable of running 2.5/2 times faster than the Pentium II, or at 375 MHz. In fact, it runs at 533 MHz. Your "constants" are starting to look a little wobbly to me, which makes your simple-Ass way of comparing processor generations by comparing operating voltages way off base.
Petz |
