CPU related questions, What and why?

[Yoad]

Hey, Although we ALL know the basics [CPU's of same series is determined hwo good their are depending on clock speed, some CPU's do more "Work" per clock cycle thus need to do less cycles in order to match other, faster clocked CPU's in performance, etc]

However, I Wonder about slightly tougher questions, I'll begin:

1. What does the Mhz mean? I know its the clock speed, but, how fast [in real time] theoretically would be 1 clock cycle that its speed is say 2166MHz[and if possible, what would be the time decrease for a cycle for each 1MHz increase in clock speed]

2. What is "Work per cycle"? In other words, what exactly is it that the CPU does while he's cycling clocks that we consider as "cpu work per cycle"? Obviusly its "processing" stuff, but what? Does it calculate equations?

3. What determines how much "work" per cycle can the CPU do? It's architecture? The Cache size?

4. How does the transistors in a CPU work? Why more is better even if they are running on slightly lower speed [due to increased power consumption and heat exauhstion]?

5 and last [this one's rather simple]. I Just wonder, how does a typicall CPU is built [in its layers, what does the nowdays CPU die has on it]?

Illustrations aswell as text explanation will be great just try to keep it simple and don't link me to messy sites full of text, im no scientist.

-Jo

 

[Apokalipse]

a clock cycle is one pulse; where each "one" or "zero" (0 = low current, 1 = high current) goes forward one stage

the higher the clock speed, the more stages the one's and zeroes are pushed forward per second

work per clock cycle: how many instructions are finished per pulse

instructions (a set of ones and zeroes - most programs are 32-bit, ie use 32 1's and 0's per instruction) go through the transistors; the transistors act depending on what two inputs they get, and what type of transistor it is

for example, there are AND, OR, XOR, NAND, NOR, XNOR
each transistor responds differently according to the two inputs they recieve

AND transistors:
input_output
0, 0 -> 0
1, 0 -> 0
0, 1 -> 0
1, 1 -> 1


OR transistors:
0, 0 -> 0
1, 0 -> 1
0, 1 -> 1
1, 1 -> 1

XOR:
0, 0 -> 0
1, 0 -> 1
0, 1 -> 1
1, 1 -> 0

NAND:
0, 0 -> 1
1, 0 -> 1
0, 1 -> 1
1, 1 -> 0

NOR:
0, 0 -> 1
1, 0 -> 0
0, 1 -> 0
1, 1 -> 0

XNOR:
0, 0 -> 1
1, 0 -> 0
0, 1 -> 0
1, 1 -> 1

there's also gates, with one input

NOT:
0 -> 1
1 -> 0

BUF:
0 -> 0
1 -> 1

 

[Meithan]

Haha, those aren't easy questions to answer! However I'll try my best, and maybe with my reply and apokalipse's, you'll have a better view of these matters.

1) This one is easy. One Hertz means simply one cycle per second. So naturally a Mhz is a million cycles per second. It's then obvious that at 1Mhz, each cycle takes 0.000001 seconds (one millionth of a second, or one microsecond). So if your CPU has a clock speed of 2166MHz, it means each cycle takes 0.0000000005 seconds (that's 0.5 nanoseconds). Increasing the speed by 1Mhz wouldn't make that much of a difference, since it's already running very fast.

2) When a CPU is "processing" information it's doing four basic things in this particular order: READS an instruction from the memory, DECODES the instruction so that it can understand it, EXECUTES the instruction on the data in the memory, and finally WRITES the resulting data in the memory. The clock of the CPU just makes sure this sequence of steps is carried out one by one.

3) This is obviously a very complex question. If it were simple, then CPU makers such as Intel and AMD wouldn't be working so hard to produce better and better CPUs. But I think that with the new Conroe CPUs, Intel has proved that a good architecture is what can potentially give the best performance. Cache is "internal memory" of the CPU, very fast memory at that, but as has been shown by recent benchmarks, it doesn't substantially affects performance.

4) The transistors inside a CPU aren't like the ordinary transistors you can buy at your local electronics shop (these are called bipolar junction transistors). CPUs use a special kind of transistor called MOSFET, which stands for Metal Oxyde Semiconductor Field Effect Transistor. How a transistor works, and particularly, how the Field Effect Transistor works, would be very hard to explain unless you're initiated in electronics (and even solid-state physics).

What you need to know, is that transistors allow us to build logic gates, which are tiny pieces of electronics that can simulate logic functions: AND, OR, NOT, etc, like those apokalipse posted. Transistors also allow to build circuits that can "read" or "write" data from a certain place (like memory).

FETs can be made much smaller than common transistors, so this is why they are used in CPUs. The smaller these FETs are, the better, since they will use less power and more can be packed on the chip die. Current CPU technology is based on a 65nm process, which means the smallest structures inside a CPU are aprox 65 nanometers in size!

How small is a nanometer? Neurons are typically around 50 nanometers across, the AIDS virus is around 120 nm in diameter, a human red blood cell is typically 6000-8000 nm in diameter, and a human hair typically is 80000 nm in diameter. (from Wiki)

5) How it's built? That I wouldn't really know. I believe it's just a semiconductor wafer on which electronic structures (yes, ultimately made out of transistors) are inserted. When zoomed-in, a CPU looks like a city, each "building" performing some specific function. Here's a pic of the die of a CPU (Athlon64) with some details:



Whew!! That was long! I hope that I was clear enough. These are indeed difficult questions.

 

[jonny_uk]

Haha doesnt ask for much does he!! LOL

 

[COD2_fanatic]

quite interesting in my opinion

 

[killians45]

Here's a little info, usually procs are made in a batch from a specific die. This die is then tested via a heat index to find out the stability point (heat/frequency/stability). This batch is then marked at a frequency that is found when the heat index reaches its maximum frequency where it can run stable under normal conditions, which is usually under par from what is actually considered normal for less heat damaged returns. Thus allowing "over clocking", but each manufacturer is different when it comes to what is allowed and what is not. Thats why some procs are move over clockable than others.