M1 max And Other Processors

[Hasan123]

As seen on the launch event yesterday, the M1 pro and M1 Max look bigger than its previous M1.If they Can get such a performance boost by just (maybe) making it "Big", then why can't Intel or AMD make their chips bigger? I don't have any idea how a chip works or how it is made. Just curious of what would happen if other chip also made bigger chips. 
Please don't laugh at my knowledge.

[LAwLz]

I think you are under the misconception that bigger = better. Things aren't that simple.

 

Let's think of cars for a moment. Super and hyper cars are usually very wide and low. Here is an example:

 

However, that does not mean that the lower and wider car you make, the faster it will go.

You also have to consider the engine, the aerodynamics, etc. If it was as simple as "lower and wider = faster" then the fastest car in the world would be a piece of paper.

 

 

 

One of the big things Apple did to make the M1 faster was going from a 4 core CPU to an 8 core CPU. But both Intel and AMD already have 8 core processors. At some point adding more cores just becomes an issue. I guess you could say both AMD and Intel have already done the "make the chip bigger" trick, and it's a pretty hard trick to do again with our current manufacturing processes.

 

 

If we stick with the car analogy, how you design the roads will affect how quickly a car can go from let's say your house to the local pizza place. It doesn't matter if you got a super car if you get stuck in traffic or have to take really long detours (in processors, this is the design of the interconnect, how quickly each CPU core can communicate with each other as well as things like RAM).

 

Then there is the issue of the tire (or in the case of computers, software). You can put however many horsepower you want in the engine but if your car has bicycle tires it won't be able to actually transfer all that power into movement. Same thing with software. If your software isn't capable of using more than let's say one core then adding more cores won't affect performance. You could say "just make each individual core bigger" and that's fair, but that's also really hard and might not scale as well as you think. Which leads me to...

 

The issue of gas. If we just keep putting in a bigger and bigger engine, the noise, weight, fuel consumption and other factors will sooner or later each a breaking point. You can't put a tank engine inside a Ford Focus because other parts of the car, such as the chassis and fuel tank are not designed for it. What good is a massive engine if your fuel tank runs out in 5 minutes? Do you really want to drive car which sounds like a jet engine? A car with an engine so big it blocks your windscreen?

Same deal with processors. If you just keep making the cores bigger and bigger, you will run into issues with power consumption and thermals. Let's for example say Intel made their processor cores twice as large. That might result in a 60% performance increase for each core. But now they are no longer able to fit 8 of them on a single chip. All of a sudden they have to go down to a quad core, or else it will use too much power, it will be too expensive to make, and it will cook itself from all the heat it generates.

Sure, a 60% single core performance increase might sound nice, but is that a good tradeoff when compared to the 100% performance increase they could have gotten by making more cores instead?

 

 

Processor design is a balancing act, and because Apple has gone about processor design in a completely different way, they get different parameters they can play around with.

For example, Intel and AMD has to design their processors around the idea of having DDR4 memory. So making massive CPU cores that need lots of memory access will just be wasteful because they will be bottlenecked.

Apple on the other hand just went "oh, our cores are too big and gets bandwidth starved by DDR4? Then let's go with LPDDR5. It's still not enough? Then we'll just have to design our own PCB and put the memory straight onto the same die, with massive 8-channel interfaces.

And that's just one example of how Apple does things very differently. Not just because Apple doesn't have to play nicely with others (like AMD and Intel that needs to appease motherboard manufacturers, memory makers and DIY builders) but also because of things like them using a completely different instruction set for their processors.

 

[Chris Pratt]

The biggest problem with big chips is that they require more and better quality wafer. In the fabrication process, there's always imperfections, so a certain amount of wafer is always wasted. The more contiguous wafer required by the chip die, the greater likelihood that there will be imperfections in that area that make it unsuitable for the chip, leading to more waste and far higher manufacturing costs, not to mention less availability since wafer is at a bit of a premium right now.