eli5 – Why are ‘System on a Chip’ computer’s potentially faster?


eli5 – Why are ‘System on a Chip’ computer’s potentially faster?

In: 0

The electrons don’t need to travel as far.

Think of the traces on a circuit board as wires. The longer the wire, the longer it takes a signal to move from one end to the other.

SoCs don’t require nearly as long a stretch of wire to get from A to B.

Yes, electrons functionally travel at lightspeed, but in a system running at billions of cycles per second, that extra time for a signal to travel a wire starts to matter.

The parts of a computer that connect the different chips generally work slower than the internals of the chips. If you put all of those parts onto the same chip as the CPU all of the interconnections are much shorter and can work more quickly. It will also likely use less power as well.

The downside is that the chip will be bigger and harder to cool. If you are already trying to cram as much as possible onto your chips, like multi-core cpus, you get to the point where it’s better to use several connected chips instead.

I suspect that within 10 years or so it will become feasible to cram enough computing power for pretty much any desktop use onto the one chip and computers will all shrink down to the size of a cellphone. We’re just about there already.

They’re not necessarily faster, that’s entirely dependent upon the components you put into your SoC, and will never match the high power discrete chips due to heat limitations

SoCs are more power and space efficient because its all in one module. There’s one set of power management circuitry and only one physical thing placed on the board. This matters for things like phones where you’re trying to minimize standby power to eek a bit more life out of the battery and minimize space so you can fit in more battery.

They can be a bit faster than expected because they can have custom interconnects between the chips. If your PC has a separate CPU and GPU then they have to talk through a PCI-e connection because that’s the standard. If you put them together in the same package then you can have them talk through a custom standard that can support more data and let you eek out a bit more performance

In the end though, a SoC will never out perform a discrete CPU and GPU because it just can’t handle the heat created by putting multiples of the really big chips together in a package. Heat is actually what limits the peak performance we can get out of chips these days

Where did you see the claim that a SoC computer is potentially faster? SoCs were introduced to save space and packaging costs not increase performance. Waste heat is the limiting factor to maximize performance these days. Cramming CPU, GPU, and other components on the same chip only makes it harder to remove heat.

A SoC can be optimized to produce the least heat per watt. In a data center where cooling costs matter that could be a decisive factor. For desktop computers it doesn’t matter. In case of a laptop it is possible to design it with discrete CPU and GPU that will maximize performance while grid powered but switch to power saving mode while on battery. But if you want to optimize size, cost, performance, and battery life a SoC can be the best solution.