3 ms·

I disagree regarding the importance of this kind of "tricks". You are right in pointing to all the "boring" stuff that is considered state of the art in chip de

I disagree regarding the importance of this kind of "tricks". You are right in pointing to all the "boring" stuff that is considered state of the art in chip design being the main reason why these chips are impressive. Apple seems to have pulled all the right strings in that regard. But there is a point at which you can't execute the known optimizations any more perfect than you already do - when you're already using the best available lithography, implementing all the well-known architectural benefits like big/little, big caches, wide decoders etc., you eventually hit a brick wall and the only thing that will get you further are new and creative tricks like the ones discussed in the linked tweets. Each may just add a little, but together they add up to a noticeable improvement with regard to a competitor who also executed all the standard stuff well, but maybe did not think about these particular new optimizations, or might not even be able to implement them due to not being in control of the entire vertical ecosystem. That "competitor" btw is not Intel - they already fail at the basics - but AMD.

saagarjha 6y ago
The point is that in benchmarks these things don’t really show up at all and Apple still leads. The things I mentioned are only useful because of specific usecases, and saying that Apple is ahead because of those is inaccurate. The point is that the tricks are extremely important for Apple, but not in an argument of “why is Apple making fast processors”. Regarding a comparison with AMD: I think we’ll have to see what Apple’s professional chips can do. My guess is they’re pulling ahead already.
- Slartie 6y ago
  So JavaScript is not benchmarked, fast ref counting does not improve native application benchmark results, and the memory model trick does nothing to improve benchmark results of x86 applications? Yeah, you can of course say that each of these is a "specific use case". But which use cases other than running native MacOS apps built on Apple's frameworks, x86 MacOS apps and web applications do you have in mind? I see those covering a lot of what people tend to do with their MacBooks.

saagarjha 6y ago

The point is that in benchmarks these things don’t really show up at all and Apple still leads. The things I mentioned are only useful because of specific usecases, and saying that Apple is ahead because of those is inaccurate. The point is that the tricks are extremely important for Apple, but not in an argument of “why is Apple making fast processors”. Regarding a comparison with AMD: I think we’ll have to see what Apple’s professional chips can do. My guess is they’re pulling ahead already.

Slartie 6y ago
So JavaScript is not benchmarked, fast ref counting does not improve native application benchmark results, and the memory model trick does nothing to improve benchmark results of x86 applications? Yeah, you can of course say that each of these is a "specific use case". But which use cases other than running native MacOS apps built on Apple's frameworks, x86 MacOS apps and web applications do you have in mind? I see those covering a lot of what people tend to do with their MacBooks.