Comment by ohgodplsno | Hacker News Reader

Comment by ohgodplsno | Hacker News Reader

Parent Full thread

> They capitalized on Intel/AMD sleeping on RISC architecture and going to smaller nanometers.

Apple is singlehandedly using up almost all of TSMC's 5nm processes. There are no smaller processes available anywhere in the world, and Samsung's is disappointing. Processes that have largely been driven by demand from AMD and Apple (EPYCs and Ryzens from AMD, and Apple's mobile chips). They capitalized on nothing except having fat wads of cash to slap people around with.

Additionally, it's not "being a RISC architecture" that makes the M1 a beast. Slap a 5950X on a SoC with directly available memory and GPU, and you get similar (better, even, considering current gen CPUs do get better performance already) results. "Deep integration of an entire stack makes things more performant" is not a new thing, it's merely that Intel and AMD have been upholding the contract of x86/64 that is interoperability and replaceability (well, Intel has been doing some socket fuckery for upgrades, but upgrades are still entirely possible), and Apple can afford to throw it away.

EDIT: bUt ThIs oVeRlOoKs PoWeR CoNsUmPtIoN

I swear I'm going to come out with a souped up raspberry pi and tell HN "look, it uses 1W of power for 10GFlops" and you would buy it because tHe PoWeR pEr wATt iS BeTTeR !11!!

This is the upside of Apple's distressing tendency to burn backward compatibility. As a Mac dev who chooses to support really REALLY old machines, I've experienced the downside of this, needing to keep dedicated retro machines just for building retro software on, and developing everything repeatedly on different generations of computer instead of having it just work the first time.

This is the upside: if they want to make a quantum leap, they will absolutely do so if it benefits them, and everybody's expected to just re-buy, at Apple prices (mind you, trade-ins are a thing). Apple absolutely can afford to throw away anything that impedes them from doing what they want, and in the case of the M1 processors it's serving them really well. Turns out the fat wads of cash can be used to buy useful things, and when Apple wants to, they can CHANGE THE WORRRLLLD ;)

I'm sure the PC (not necessarily 'x86') world will find ways to more or less keep up: there are plenty of people who really, really want to keep up. It's just that complacency won't cut it: PC-land has to really seriously work to not fall way behind. Good. That's best for everybody. As a lifelong Mac guy since the 68040, I'm delighted the PC world has driven Apple to these extremes, because now I get to enjoy the results.

The 5nm process is only one part of what makes the M-series so fast and efficient. Yes, RISC is another but much of Intel’s chips are RISC at some levels.

A big part of what makes Apple’s chips so fast is that they optimize them for the needs of the OS. In interviews they have given examples of how to profiled the common operations in the OS and then made sure that those operations were as fast as possible on the chips. that integration is a big part of it. Other hardware and software vendors are not as tightly coupled and don’t do this kind of optimization.

Do you own a laptop? Is the ability to burn your lap when compiling a feature for you?

> when compiling

When opening a second chrome tab*

That might compete on performance but I wonder how “easy” it would be to compete on the power-performance curve.

It's possible, yet noone is doing that. I wish someone did!

The Udoo Bolt v8 is about the tiniest x86 "SoC" that I know that runs a Ryzen V1605B, 16GB of RAM and a Vega 8 that costs $500 and scores about half of the M1 Max in Passmark. Not too shabby for something that still has to deal with DDR4-2400 and storage far away!

This overlooks power consumption

Reply on news.ycombinator.com