Why are CPU and GPU manufacturers trying to make transistors smaller and smaller instead of making the chips bigger so they can put more transistors?

318 views

Why are CPU and GPU manufacturers trying to make transistors smaller and smaller instead of making the chips bigger so they can put more transistors?

In: 11

13 Answers

Anonymous 0 Comments

Costs and distance.

The closer 2 parts of physically the faster it can be which is something people forget.

Anonymous 0 Comments

bigger chips would mean the sizes of certain other parts in a PC needs to be changed as well.

i already had a hard time when switching my 1060 to a 2080 due to their size difference

Anonymous 0 Comments

While some comments touch on the topic of distance and latency, I really want to press the importance of this point. If you get too large then the latency gets to the point where it’s difficult to synchronise across the chip. If things aren’t in sync then it’s not going to function.

Anonymous 0 Comments

We’re talking square laws here. Twice the size is 4 times the cost.

Also, on average the size of the transistors gets reduced with about the square root of 2 each generation.

Anonymous 0 Comments

bigger chips is not cost effective for manufacturers, as these thnigs are produced in Silicon wafers which are individually expensive. so there is a economical interest in fitting as many chips as possible per Unit.

this also falls into the 2nd part: Yield, some chips will be faulty at production, a smaller chip means a faulty unit wont be as much of a loss.

Anonymous 0 Comments

There are two major limitations to device speed – capacitance and distance. When a transistor turns on, the signal it generates has to reach the next transistor. First, current has to flow through the transistor to charge up the capacitance of the link to that transistor, and the next transistor – so the smaller, the better – and the signal has to travel the length of the link – so the shorter the better.

By making transformers smaller, the links become both smaller and shorter, so devices can work faster.

Anonymous 0 Comments

On top of the other answers here look at Moore’s law. Transistor counts double every two years. If you double the size of your chip you’re really not jumping really far ahead of the curve in terms of transistor count and you’ve doubled the footprint of your chip. Do it again and now your chip is four times the footprint and all for minimal gain

Anonymous 0 Comments

The speed of light limits the size of chips. With a 3 Ghz processor, light travels about an inch per cycle (electricity in silicon moves a bit slower than that). Anything that has to happen in one cycle can’t have a longer path than that.

You can get around this by putting in multiple cores, to an extent. You can split up a program so that each core handles separate tasks (threading), but there’s diminishing returns because they can’t share the tasks. If your program has 3 tasks, with one task taking 50 cycles and the others taking 25 cycles, then with one core it takes 100 cycles, with two cores it takes 50 cycles, with three cores it takes 50 cycles, and with 16 cores it still takes 50 cycles.

Anonymous 0 Comments

So you have a limited amount of space. With space not being infinite each application would in theory have a maximum volume that the GPU and CPU can occupy. No one wants a Computer that takes up a full room anymore. So how do you increase the amount of something you want when you have a space constraint? You increase its density. So by shrinking the size of the transistors now we can fit more into the same space or put the same amount of into a smaller space.
If you do want to build a computer of colossal size it will now be able to be more powerful by taking advantage of the faster chipsets or (hopefully) less expensive by taking advantage of the older and less performance dense chips.

Anonymous 0 Comments

In addition to other reasons, smaller transistors have less capacitance and are therefore faster. That means more bandwidth in a communications application and more processing power in computers.