So this is probably really stupid, but why, for a company such as AMD, cant they make a CPU similar to the 3970x but with 8 or so big cores with improved performance? I keep seeing insane core counts from them, but I don’t quite get why they can’t make really good single-core performance.
In: Technology
We are at the point where a single CPU core can’t really get significantly better without also producing so much heat as to basically vaporize itself – heat production being *very* nonlinear with performance – remember that for extreme overclocking to 6GHz and above people have to constantly dump liquid nitrogen on CPUs as it boils off. We need to invent new materials/technologies to step forward.
On the other hand getting a bunch of cores and dumping them on one chip is a lot easier and results in actual tangible performance improvement, with *much* less heat output increase than trying to get the same processing power into one core.
If one has a big house and multiple cabinets and kitchen appliances but only one person (thread) then that takes time to process serially in length versus a team of people (multiple threads).
A small house and kitchen cannot fit many people walking about either, which goes to the physical traces of people routing to needed parts of the house.
Then there is the agreed layout and remembered organization of tools and ingredient access (data storage and memory addressing). If there is any conflict of calls or needed slowdown to translate, this adds time to resolve for performance access of a timely manner.
In short, this is a balance of the intended output of a quick bachelor meal or a banquet for a large family gathering. Larger houses can process more at the cost of energy and heat with the space and usage.
To merge cores would require to decide what appliances and house layout to keep for optimal energy use with little uncomfortable heat for the people that walk about the area and this is given the output intended (a simple dedicated meal versus an excessive large buffet of varying side dishes).
There’s a whole host of reasons, heat production, physical limitation to addressing, bandwidth to each core, etc.
Basically, if each core has its own addressing, it gets it’s own data pipes, its own RAM, etc. One giant core would still be limited to what a single core could address, you’d also run into limits of thermodynamics. Power consumption would be off the scale, and there would be bottlenecks all over the place.
Bigger is not better for CPU cores. Think of a CPU core as an engine that will always output the same amount of power regardless of its size (More on this in a bit). But the bigger it is, the more heat it generates. To prevent it from melting itself, we need to cool it. You cool it by either sticking something on it to suck away the heat or by slowing it down. Smaller CPU produces less heat which means it can run faster for longer.
Now on the (counter) intuitive nature of bigger does not equal more power. A CPU is a logic circuit. Which means when you send electricity in a certain pattern through it, it will send electricity out in a certain pattern. Power for a CPU is determined by how quickly it will send that pattern out. Suddenly doubling the number of transistors in the CPU core is not going to make it spit out that pattern any faster. It depends on how quickly electricity can get in and out of the core, not the number of transistors it goes trough to do it.
So why have we been shoving more cores into CPUs? Well, the chip is going to be a certain size. Transistors have been shrinking in size for as long as they have existed and they can fit more transistors onto the chip. We’ve long passed the time where we really need to add more transistors to the CPU core to make it faster. So the next best thing to do with all that extra chip space is to shove more cores in there (Technically cache space helps but that’s another story).
You’ve mentioned stuff like TurboBoost that Intel has. Well, in that scenario Intel does a clever trick because their CPU is limited by heat. The CPU itself can actually go really fast (This applies to AMD as well), much faster than their quoted specs. But the problem is that running it at that speed will basically melt the CPU. So they shut down unneeded cores which now no longer produce heat. That means the one running core can now crank up its speed because it can produce more heat by itself before melting the whole CPU.
Latest Answers