The Relationship Between CPU Frequency And Performance. Critique, Also Answer Qs?

[Excelsior]

All things being equal, shouldn't the answer be "never"?

It's just another factor of performance. It should never hurt to take increase it, but real-world issues like heat give you a short-term limit. But assuming heat isn't a problem going to a certain clock speed, then it is always going to be a good thing because IPC and bandwidth of the various components of the system won't suffer for it.

Adding more mhz can eventually reach the physical limits of a chip, as in the speed of electricity isn't fast enough, not enough bandwidth from the cpu to the motherbaord to complete operations and store int he RAM, even i the ram is fast enough etc. Also, there may be a point where the program executes XXX in near 0 seconds, and adding more Frequency doesn't help, which if the other components aren't a bottleneck then the software's code itself can be the bottleneck.

I think you can only extrapolate this if you can divine an equation expressing the performance of the cpu. The fact that it is sitting in a system that has a hard drive and ram and things like that in it, and that you are running benchmarks that might get bottlenecked by those things, doesn't help.

As I said, It'd be possible to find how much a slow hard drive hinders, XXX ram hinders, etcetera. Obviously you can take them into account. This may not be accurate on a hundred ghz jump, but at a 1.4 ghz jump my results remained within a few percentage points. If I had booted certain linear benchmarks at 1000 mhz, and then 1500 mhz, I'd be able to ascertain the performance at 2400 mhz, then actually try it.

 

[Quailane]

Adding more mhz can eventually reach the physical limits of a chip, as in the speed of electricity isn't fast enough, not enough bandwidth from the cpu to the motherbaord to complete operations and store int he RAM, even i the ram is fast enough etc. Also, there may be a point where the program executes XXX in near 0 seconds, and adding more Frequency doesn't help, which if the other components aren't a bottleneck then the software's code itself can be the bottleneck.

Well, what I was saying was that the memory can't feed stuff to the processor fast enough, so it is a bottleneck. At what cpu speed is it using all of the data from the ram so fast that adding any extra speed does nothing.

 

[XWRed1]

Adding more mhz can eventually reach the physical limits of a chip, as in the speed of electricity isn't fast enough, not enough bandwidth from the cpu to the motherbaord to complete operations and store int he RAM, even i the ram is fast enough etc.

Those are all either implementation details or bottlenecks from other subsystems.

Being bottlenecked going to secondary storage is going to kill your performance if you chose to emphasize IPC too. I'll go out on a limb and speculate that "speed of electricity" would be a problem for any aspect of the cpu's design, if you got that far with it.

At some point the cpu still has to do work... it has to move electrons around for that to happen, it is unavoidable.

At what cpu speed is it using all of the data from the ram so fast that adding any extra speed does nothing.

The point at which you don't care if your program ran any faster?

 

[Excelsior]

So I'm going to present my paper tomorrow. Anyone have any last minute suggetsions on things to change or such?