Clock, Hertz and IPC

[Guest]

Hi,
Sorry if I'm in the wrong place for asking this,
I'm not that beginner but want to know more about main Chip of the Computer; CPU.
My main Question is that What exactly the IPC is?, and why a CPU clock rate of for example 3.2 GHz may have smaller power than a CPU with a 2.8 GHz?
And what exactly the Clock of CPU is and so on?
Maybe you could call it 'CPU Specs 101' 😀
Thanks.

 

[hunter315]

IPC - Instructions Per Cycle
Hertz[Hz] - Cycles per second

The clock speed of a CPU is always the large number of Hz that is listed, like a PII 955 @ 3.2GHz or an i7 920 @ 2.66GHz.

Its difficult to get exact numbers for the IPCs but the intel i cores are slightly faster than the AMD Phenom II cores, so if the AMD cores have an IPC of 1(for convenience of math, not anywhere near true) and the Intel i7 cores have an IPC of 1.1(so 10% faster, again for easy math) then the AMD core @ 3.2GHz executes 3.2 billion instructions per second, and an Intel i7 950 @ 3.03GHz will execute 3.333 billion instructions per second, the speed of a CPU is far more about how it is laid out(which you cannot tell from specs) than what its specs are, sometimes a "slower" chip(clock speed) will be faster(performance) because it has a better internal layout, or vice versa.

 

[Timop]

The clock is measured in Hertz, so the 3.2Ghz or 2.8Ghz would be a CPU's clock.

IPC stands for "Instruction per clock", or in simpler terms, how much work can a CPU do per Ghz/Hz/etc. A CPU with a higher IPC can do more work, or be more powerful, than a CPU with a lower IPC if both are clocked the same.

An example would be how a 3Ghz Core i5 Quad can beat a 3 Ghz Phenom II quad, as the i5 has a design with a higher IPC.

 

[Guest]

Thanks.
So we could not measure exactly a CPU's IPC?
why MIPS does not use by companies these days?
and if there is, which programs could tell us about a Processor's IPC?
Thanx again!

 

[hunter315]

MIPS and IPCs arent measured and posted because they arent possible to standardize, if AMD uses an integer add instruction for their MIPS calculations and Intel uses a floating point multiply then the AMD MIPS will be much much larger, but if they both measured with the same instruction, say a floating point addition, then they would be much closer. They would have to give you the IPCs for each instruction in their set(there are a lot of them) and you would need to know the breakdown of the program you are using to be able to make sense of it. Its wayyyy too much math, and doesnt tell you much more than X will run A faster than Y can run A, which we use benchmark programs to do instead of doing lots of math.