Indicators of performance and what they mean

I'm aware of some metrics of hardware performance but I'm not sure what physical aspect increases these values, or what it means in real world performance.

CPU: MIPS, MFLOPS, Cache Latency, Cache Bandwidth, Inner Core Bandwidth, Inner Core Latency.

How does Intel/AMD get more instructions/clock and floating-point-operations/clock? What do you need more instuctions/clock for? I assume cache/core latency and bandwidth is a function of clock speed and shorter lengths of the connections; what applications would an increased value of this benefit?

I have a little bit of a better understanding on GPU performance.

GPU: Vertex processing rate, Pixel processing rate, Texel fill rate, Pixel fill rate, Memory Bandwidth.

This is what I think is true:

Vertex processing rate determines how fast the processor can compute vertex information in 3D space to create a wireframe, the information is in (X,Y,Z) form. Pixel processing rate determines how fast the processor can computer pixel information (the single dot on a screen), the information is in (R, G, B, T) form. Pixel fill rate is how fast the pixels can be transfered from the processor to the monitor (I don't think you need more than your resolution size x FPS). I'm not at all sure what texel fill rate is, but I know a texel is a texture unit (just not sure what form the information is in). Video memory bandwidth is how fast completed images are transfered from the processor to memory banks and vice versa.

I'm pretty sure a GPU can do faster math calculations than a CPU, so why aren't most math calculations done in the GPU?

i can help you out with a few of these:

Based on the in-built algorythms inside the chip (pre-defined calculations) a processor can make-lite of certain mathematical forumulae - SSE, SSE2, 3DNOW, etc... instructions are all a bunch of mathematical equations that are common for certain programs. By programming these equations inside the chip, the computer can find the result a lot quicker, especially when done at 3000 cycles per second.

A chip will be faster if it has more equations to use. So Intel can make a faster chip than AMD because of these in-built equations, even though they are the same clock speed because the intel chip can work the results out faster.

If you can get more instructions per cycle, then your processor is doing more things per cycle. To elaborate, a 3.0GHz processor can do 3000 instructions per second. If you can get your processor to do more, or make these instructions easier by using in-built equations, you make your computer faster. for example. If you wanted to work out the radius of a circle. A computer would have to work it out the long way. WE know how to use the equations related to PIE. If you TOLD your computer all about PIE, and gave it the equations, it would not need to work it out the long hard way, all it would do is use those simple equations. There would be less requests for data, less calculations, and as a result, a quicker computer... if you can see what i mean. its quite hard to explain.

A GPU only has instructions for 2D/3D applications. So it would not be very effective at processing anything else, because its in-built instruction set is not defined for doig anything else. Its like telling a scientist to build a brick wall. He could probably do it, but nowhere near as fast as a brick-layer could, because the bricklayer knows all the short-cuts, etc....