Take the 2-minute tour ×
Super User is a question and answer site for computer enthusiasts and power users. It's 100% free, no registration required.

In PC hardware, such as CPUs and GPUs, is the transistor count 'actual' or is it like saying this car has 300 horsepower. The car doesn't really have horses under its hood. Is it the same with the chips?

If the transistor count is 'actual' how are the engineers able to manipulate more than 4 billion transistors (and counting) every year or so for new architectures? Do they actually manipulate individual transistors (I know they don't interact with it physically, but when designing the architecture, do they actually fiddle around with individual transistors on the schematics?)

share|improve this question
1  
Horsepower is a well-defined unit of measurement and doesn't even try to imply that there are horses under the hood. Both horsepower and transistor count are quantifiable, measurable units. A car that says it produces 300 HP really does, otherwise the manufacturer could be in trouble. Same thing with transistor count. CPUs and GPUs are built by assembling and integrating blocks of functionality that are well-defined. The design tools know exactly how many transistors (and gates) each block uses. –  Dan7119 Jan 12 '12 at 19:16
    
@Dan7119: I understand that they are a unit of measurement but we don't say that a car that has 300 HP has 300 horses in it. I myself believed that the transistor count is actual count, but got into an argument with someone who gave me the HP analogy and said that "a GPU with 4+ billion transistors is like saying a car has 300 HP. It doesn't really have 4+ billion transistors but acts like it does as far as computation power is concerned". –  Samaursa Jan 13 '12 at 3:58

2 Answers 2

up vote 9 down vote accepted

A transistor is an actual, discrete, quantifiable unit, composed of P-N junctions, substrate junctions, and small insulating layers. They normally aren't manipulated individually, but rather are added in larger functional blocks such as SRAM arrays, ALUs, etc., although sometimes timing issues may need to be resolved by individual manipulation.

share|improve this answer

The chip really does have transistors on it, but at the same time, this might not be a good measurement of how quick the chip is.

A transistor is the basic unit that is required to make NAND, NOR, and inverter gates, which are then used in combination to make everything from adders, to memory units, to muxes to determine which instruction's result should be returned.

Generally speaking, it is not useful to manipulate any unit smaller than a NAND (4 transistors), NOR (4 transistors), or inverter (2 transistors) when designing a chip, as all structures on the chip are combinations of these units.

Edit: There are cases where you will only use one transistor, primarily in multiplexers and latches, but they are not strictly necessary (though they are more efficient, and sometimes used). Thank you Dietrich Epp for the update.

share|improve this answer
    
Most chips are not made only from NAND, NOR, and inverter blocks. You can, but the blocks are usually larger (or just different, like latches). You can download block libraries from ASIC manufacturers if you want to see a sample of what blocks chips are made out of. A functional block (like a small multiplexer or adder) is made directly from transistors and then synthesized into ASICs. You can bet that the folks at Intel don't work this way all the time, sometimes making larger components directly from transistors. –  Dietrich Epp Jan 8 '12 at 9:58
    
Latches are made from those units as well. Are you aware of any use for just one transistor? (any units made from something other than NAND, NOR, and inverters) –  soandos Jan 8 '12 at 10:00
    
Latches can be made from those components, but they often are not. –  Dietrich Epp Jan 8 '12 at 10:00
    
So how are they made (this is the standard implementation that I have seen everywhere)? –  soandos Jan 8 '12 at 10:02
    
They teach you that example in school because you can build everything out of gates, and the gates are easier to understand than transistors which only pull in one direction. Here is an example of a CMOS latch: iiith.vlab.co.in/?sub=21&brch=66&sim=519&cnt=1 Note that RS latches are not very common in synchronous circuits, you're more likely to see D-type latches. It would take several additional transistors if you wanted to make a D-type latch out of gates instead of out of transistors. They don't teach D-type latches in intro classes because synchronous circuits are hard. –  Dietrich Epp Jan 8 '12 at 10:09

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.