Why aren't CPUs bigger?

Why aren't CPUs bigger? [closed]


From today's perspective, this question does not fit our Q&A format. We expect the answers to be supported by fact, reference, or expertise, but this question is likely to spark debate, argument, vote, or expanded discussion. If you think this question can be improved and possibly reopened, get help.
Closed 8 years ago.

CPUs are relatively small, and engineers are constantly striving to downsize them and get more transistors on the same surface.

Why aren't CPUs bigger? If an approx. 260 mm 2 Chip can hold 758 million transistors (AMD Phenom II x4 955). Then a 520mm 2 can accommodate twice the number of transistors and technically twice the clock rate or cores. Why isn't that done?






Reply:


Basically you are right: increasing parallelization is not only feasible in the short term, but also that only one Path. In fact, you propose multi-cores as well as caches, pipelining and hyper-threading: Speed ​​gains through increased use of the chip area. Of course, shrinking geometries do not collide with the increasing utilization of the nozzle area. However, the yield is a major limiting factor.

The yield increases in inverse proportion to the chip size: large chips are more likely to "catch" wafer defects. If a wafer flaw hits a die, you can throw it away. The yield obviously affects the cost. So there is an optimal tool size in terms of cost and profit per tool.

The only way to produce significantly larger chips is to incorporate fault-tolerant and redundant structures. This is what Intel is trying to do in its Terra Scale project (UPDATE: and what is already practiced in everyday products, as Dan points out).




There are a lot of technical concerns (path lengths become too long and you lose efficiency, electrical interference causes noise) but the main reason is simple that many transistors would be too hot to cool sufficiently . That’s the whole reason why they’re so keen to downsize the die - this allows for performance gains with the same thermal values.







Some of the answers given here are good answers. There are technical issues with increasing the size of the CPU and there will be a lot more heat to deal with. However, they can all be overcome with sufficient incentives.

I would like to add what I think is a key issue: economy . CPUs are manufactured in wafers, as is the case with a large number of CPUs per wafer. The actual manufacturing costs are per wafer. So if you double the area of ​​a CPU, you can only put half as many on a wafer, so the price per CPU doubles. Also, the entire wafer does not always come out perfectly, errors can occur. So if you double the area, the chance of a defect in a given CPU doubles.

From an economic point of view, the reason smaller and smaller things are being made is to get better performance / mm², which is the deciding factor for value for money.

TL; DR: In addition to the other reasons mentioned, doubling the CPU area more than doubles the cost.



Adding more transistors to a processor does not automatically speed it up.

Increased path length == slower clock rate.
Adding more transistors increases the path length. Any increase must be put to good use as it increases costs, heat and energy, but decreases performance.

You can of course always add more cores. Why don't you do that? Well they do.




Your general assumption is wrong. A CPU with a double the size of a die doesn't mean it can run at double speed. This would just add more storage space for adding more cores (see some Intel manycore chips with 32 or 64 cores) or larger caches. However, most current software cannot use more than 2 cores.

Therefore, the increased die size increases the price massively without generating a profit of the same amount. This is one of the (simplified) reasons CPUs are the way they are.



In electronics, SMALLER = FASTER 3 GHz must be much smaller than 20 MHz. The larger the connections, the larger the ESR and the slower the speed.

Doubling the number of transistors does not double the clock rate.




The cost of making the raw wafers is a factor. Single crystal silicon is not free and the refining process is a bit expensive. So if you use more of your raw material, the costs go up.


Size Creature , artificial or not, like dinosaurs, are losers. The relationship Area / volume is not fair for their survival: Too many restrictions on energy - any form - come in and out.


Think of a CPU as a network of connected nodes (transistors). To provide more functionality, the number of nodes and the paths between them are increased to some extent, but this increase is linear. So one generation of a CPU can have a million nodes, the next 1.5 million. With the miniaturization of the circuit, the number of nodes and paths is compressed into a smaller footprint. The current manufacturing processes are up to 30 nanometers.

For example, suppose you need five units per node and five units of space between two nodes. End to end you can create a bus with 22222 nodes 1 cm apart in a straight line. You can make a matrix of 493 million nodes in a square CM. The design of the circuit incorporates the logic of the CPU. Doubling the space does not increase the speed, it would just allow the circuit to have more logical operators. Or in the case of multi-core CPUs, so that the circuit can do more work in parallel. Increasing the base area would actually reduce the clock rate, as the electrons would have to travel longer distances through the circuit.

We use cookies and other tracking technologies to improve your browsing experience on our website, to show you personalized content and targeted ads, to analyze our website traffic, and to understand where our visitors are coming from.

By continuing, you consent to our use of cookies and other tracking technologies and affirm you're at least 16 years old or have consent from a parent or guardian.

You can read details in our Cookie policy and Privacy policy.