Can any help me understand CPUS more?

Many reasons, mostly quite technical. One is the AMD processors share a single floating point unit for every two cores. A floating point unit is a large circuit that multiplies two numbers like 2.34333 E-4 X 4.5599 E12. Normally a CPU will break up that multiplication problem into pieces and it takes forever. The Floating point units are great, but AMD has to share them.

Another problem, ironically, is the on-chip RAM controller. Years ago, the RAM of a computer was controlled by a special chip seperate from the CPU. This is for technical reasons becuase RAM (random access memory) is actually DRAM (Dynamic random access memory) so each section of it must be read periodically or it will erase itself. Sure it's a pain, but it can be very dense. Anyways, first AMD incorporated a RAM controller on the CPU. This gave it a huge advantage over snaily Intel. It both saved power and was quicker. Then, a few years later, Intel came out with their own on board RAM controllers. Sure, it took them a few years longer, but man, they did it right! Even now, lowly 1600 RAM is plenty fast for Intels, for they do such a masterful job of keeping the RAM fed to the CPU memory cache and writing data to the RAM.

Then you have Intel's superior process. The Ivy Bridges has minute 20nm length "fin fets" that are state of the art, while AMD is still playing with old fashoned planer 28nm processes. It's hard to explain, other than the "fin fets" leak less current and turn on much stronger than planer fets so you can optimize them better.

Ivy Bridge processors are so good, that Intel actually handicaps them. The manufacture them with cheap thermal paste so they overheat quickly, despite their low power consumption. This way they are only marginally faster than the amazing Sandy Bridge Intels, that have a masterful archetecture that left AMD in the lurch. A few guys actually took apart an Ivy Bridge processor and re-packed them with quality thermal paste. They got an additional 0.5GHz speed out of them with normal air coolers. Just think, Ivy Bridges should be running between 5.0GHz and 5.5GHz. Intel did not do this to save money, just think, they buy thermal paste by the gallon, and what would a pea sized bit of paste cost, wholesale? Cents.

It appears the writing is on the wall and AMD has already stopped competing with Intel on speed. I see nothing but a slow, painful, expensive ride to oblivion for AMD. Which is really sad. Intel will become a real monster if there is no competition. Prices will rise, quality will drop, and innovation will fail.

the reason behind the speed difference is cpu architecture and also lithography (22nm, 32nm, 45 nm etc.

Now consider a simple action like a "right click". You press the right mouse button, and a menu pops up but to make that happen your OS has to communicate with the hardware and specifically the cpu.

this communication happens through assembly level language (or simple machine codes) these codes work on the very basic/fundamental electronics principals like and/or functions or NOR and NAND functions (obviously not limited to these)

these functions are basic functions because these work at the binary level i.e 1's and 0's. having established that, lets see

Lithography: how do you achieve 1 and 0 at the hardware level? the answer is voltage. if we could have two levels of voltages, we can represent those levels by 0 and 1. Say +5V is seen as a 1 and -5V is seen as 0.

now if I want to write 101 it can be seen as +5V then -5V and again +5V right?
but how do we switch between these voltage levels? we can use transistors as switches. and the smaller a single transistor will be, the more of them can be squeezed into a given area. Lithography is the size at which these transistors are developed and as litho decreases, 45 nm to 22 nm, in the same die area, there will be more transistors with 22nm.

more transistors=more switches=mores basic operations in a given clock cycle= more execution power= higher speed  

CPU Architecture: now the 1's and 0's we saw in previous post have to be arranged in sets. with each set meaning something so that several such sets of 1 and 0 can be feed into the cpu for an output.

these are instructions. every cpu arch has an instruction set which is used to direct a cpu to compute a result. It's like cpu vocabulary. The smaller is your vocabulary, the more effort it will take to communicate something. The larger it is the better is the communication and less is the chance of error.

As architecture improves, instruction handling improves. To understand instructions, consider this hypothetical example: Compute the result of 5x5. to execute this, you need to perform 5+5+5+5+5 right?
now suppose cpu architecture allows me three variables only as A,B and C. so I can store a value in A, then store another value in B then invoke the operator "+" and store the result in C.

to perform a simple task like 5+5+5+5+5, We have to use the above idea as below
A=5, B=5, A(5)+B(5)= C(10)
"clear" A
make C(10)=A, so A=10 now
Clear C, So C is empty again

now we have
A(10)+B(5)= C(15)
.
.
.
.
so on till
A(20)+ B(5)= C(25)
STOP

So we perform three steps of clearing A, performing C=A and then clearing C before each addition cycle.
we do this thrice to get to 25 so we actually end up performing 9 operations extra other than plain addition.

Now suppose a new cpu architecture allows you 6 variables
you just perform
A=B=C=D=E=5
A+B+C+D+E=F
STOP

while we used 13-14 steps using the first cpu's logic, our second cpu allowed us to write a shorter program which will be computed with less instructions as well since you are not clearing and reloading Values in register here.

so even though both CPU's have the same clock speed, and they both execute instructions at the same speed too, second cpu will be faster simply because it has less instructions to execute.

good luck
-satyam

copy pasted part of the above from an earlier thread: http://www.tomshardware.com/forum/361905-28-core2quad-q...

Sorry, but this really isn't true in this case. The OP asked about a Phenom II, which has a single FPU per core - you're thinking of the newer Bulldozer architecture, which changed to a single shared FPU per module.



Again, not exactly. The memory controller wasn't the reason AMD was ahead in the P4 era - the bigger reason was the core microarchitecture itself. The P4 had a rather severe penalty for branch misprediction, and despite having extremely high clockspeed, it had rather poor efficiency (it was really bad at using its clockcycles efficiently). The Core architecture changed that - even though they kept a front side bus (and thus no integrated memory controller) for the first couple of generations of Core architecture, they dramatically reduced pipeline length, reducing the penalty for a branch misprediction. They also improved the prefetching algorithms, which made it very likely that anything the CPU needed would be in its cache, and they gave the CPU a lot of very fast cache, significantly reducing the impact of the slow memory controller. They have gone to an IMC now, but that's only a small part of what made modern Intel CPUs faster than their AMD couterparts.

Also, Intel is currently on a 22nm process, while AMD is on a 32nm.

They manufacture them with cheap thermal paste because it's cheap. No further conspiracy theory needed. In addition, you have to keep in mind that many modern CPU overclocks aren't thermally limited - modern CPUs are very easily damaged by excessive voltage, and many of those incredible overclocks you see are at voltages which are actually unsafe to the processor (and would damage it over time). Also, the thermal paste they use currently is instead of the old method of soldering the die to the IHS, which doesn't just save the cost of the thermal paste, it also simplifies the manufacturing process significantly. This is where most of the savings occur.

As I said before, there's no reason to invent a conspiracy theory - enthusiasts who overclock are a tiny portion of Intel's marketshare, and as a result, they wouldn't do something like this just to hurt them. Intel is mostly concerned with making chips that run fast on very little power, and making them as cheaply as possible to help bring down the costs of ordinary computers sold by OEMs (with as much profit as possible).



That doesn't necessarily follow. Prices would rise in a pure monopoly, yes, but innovation would continue and quality would stay fairly high. This is largely because if Intel decided to stop innovating, people would have no reason to buy a new computer, and thus Intel would not make any money.