How did Apple Silicon get 50% faster in three years? – Daniel Lemire's blog

Posted by ibobev 1 day ago

Counter27Comment12OpenOriginal

Comments

Comment by ksec 1 day ago

Well first A20 Pro didn't get wider, it actually shrink back to 9 decode width. It would be a lot better to compare it on the A silicon rather than M.

Second is that Apple's improvement were done when they were state of the art and leading the pack. This is a lot harder than when you are behind and being shown how much more room to grow.

The A20 Pro is the largest leap in performance in recent or perhaps in all of Apple SoC history. It doesn't get enough credit and attentions that it deserves.

Compared this to the article "How did AMD Ryzen get 50% faster in two years?" The Zen 3 uArch being used in the article came out in 2020. So it is more like AMD got 50% faster in 5 years.

And perhaps this also answer and refer back to the previous HN submission and comments as to why ARM Mac are slower on SVT-AV1 encoding. Apple's SIMD performance is still behind AVX-512. But it is for good reason though and the trade off IMO is well worth it.

Comment by MiroslavPokorny 1 day ago

How much is SIMD actually useful in general computing ? I can see it being useful for stuff like graphics, video, audio and maybe AI. For other general purpose computing like browser rendering or word processing or spreadsheets the impact will be much lower.

Comment by dsign 1 day ago

Roughly, there are two types of code in a user application: logic and media related. There are many exceptions, but logic code generally doesn’t need a lot of CPU. Media algorithms on the other hand are all CPU hogs, and SIMD helps there quite a bit.

Now, I also happen to think that modern desktops run a lot of background survtech…some from the operating system itself, but most from third-party software and web pages. So, in addition to more CPU performance, I would also like more system utilities that spy on the spies and automatically puts them in some public score of shame.

Comment by kristianp 1 day ago

The simdutf Library has shown that SIMD can be useful for text processing too. Some aspects of browser rendering have been accelerated such as pathfinder_simd in servo.

Comment by MiroslavPokorny 1 day ago

Exactly SOME, most of the other stuff where the CPU spends a lot of time is not. Most code is boring if do this else do that, its quite rare to actually process a lot of parallel data.

Comment by OrangeDelonge 1 day ago

Parsing large JSON can supposedly be much accelerated by SIMD

Comment by MiroslavPokorny 1 day ago

Well theres not much point in just parsing a large json file, there will always be more work which cant be done using SIMD and that work time will be significantly larger than the portion spent parsing.

Comment by thenthenthen 1 day ago

Answer might be in the comments on the page: Apple has access to the best TSMC nodes.

Comment by wmf 1 day ago

AMD is also using N2 so it will be interesting to compare Zen 6 against M6.

Comment by thenthenthen 22 hours ago

Yes, i am sceptical of the claim in the article

Comment by mannyv 1 day ago

Better architecture.

Apple went wide, not deep. I believe even the A series were wider than normal. Also that MMU overhead must be a killer for everyone else.

Comment by MiroslavPokorny 1 day ago

What about the MMU ?