Additional Benchmarks Reveal A15 Bionic Performance Improvements in iPhone 13 and iPhone 13 Pro

[MacBH928]

I think you’ll see 14% single core improvement. Multicore will be much more of an improvement.

This is off topic but I want to know from your expertise
Given how good Apple Silicon is, will we see similar performance chips released for Windows, Linux, Android, or no one will be able(or care to) release something similar to AS and Apple will forever excel in this area? Do you think there is a possibility someone will even exceed Apple?

 

[cmaier]

This is off topic but I want to know from your expertise
Given how good Apple Silicon is, will we see similar performance chips released for Windows, Linux, Android, or no one will be able(or care to) release something similar to AS and Apple will forever excel in this area? Do you think there is a possibility someone will even exceed Apple?

It’s always possible, but who would be motivated to do so? I wouldn’t count on Android - they’ve been trying forever. And there’s not enough volume behind Linux for someone to invest the R&D. So you’re looking at windows. But only Microsoft would have the motivation, and the capability of hacking Windows to the point where it works as smoothly as macOS did for the transition. But MS doesn’t really have the capability, so they farm it out to someone? Qualcomm and nVidia can’t do it - they’ve tried. I dunno, seems unless someone reconstitutes a design team of people who’ve been successful in the past, it won’t happen any time soon.

 

[firewood]

If Google can train DeepMind to win at Go, what are the odds that they can train a data center full of water-cooled TPUs to do world class compiler optimization + CPU micro architecture + circuit design + layout, and all at once? If so, verifying the result might be an interesting challenge.

 

[Henry 3 Dogg]

The Metal score for the iPhone 13 is significantly lower than the corresponding 14216 score for the iPhone 13 Pro due to the fact that the Pro models (and the new A15-equipped iPad mini) feature a 5-core GPU while the regular iPhone 13 models include a 4-core GPU.

So how does a 25% increase in number of graphics cores produce a 35% increase in performance ??

 

[Henry 3 Dogg]

It’s all click bait. Everyone in the YouTube or Twitter is trying to make money based on guesses, estimates and stories.

It's easier to make money by saying what Samsung will pay them to say.

 

[MacBH928]

It’s always possible, but who would be motivated to do so? I wouldn’t count on Android - they’ve been trying forever. And there’s not enough volume behind Linux for someone to invest the R&D. So you’re looking at windows. But only Microsoft would have the motivation, and the capability of hacking Windows to the point where it works as smoothly as macOS did for the transition. But MS doesn’t really have the capability, so they farm it out to someone? Qualcomm and nVidia can’t do it - they’ve tried. I dunno, seems unless someone reconstitutes a design team of people who’ve been successful in the past, it won’t happen any time soon.

Why do you say Microsoft, Qualcomm and nVidia can't do it? They have all the money in the world. What does Apple know that they don't?

If Google can train DeepMind to win at Go, what are the odds that they can train a data center full of water-cooled TPUs to do world class compiler optimization + CPU micro architecture + circuit design + layout, and all at once? If so, verifying the result might be an interesting challenge.

I am still bothered with this because my understanding is that a computer can only beat human if the outcomes are limited, and in the game of GO , unlike chess, the outcomes are unlimited so the computer can not calculate the best choice. How does this even work!? Have we really reached AI?

 

[jido]

Why do you say Microsoft, Qualcomm and nVidia can't do it? They have all the money in the world. What does Apple know that they don't?

It is not easy to design chips. A company like Arm is competent enough but world-class designers can do better, and Apple employs a few. Also Arm design objectives are different which can come into play (they aim to serve a variety of use cases not just phone chips.)

I am still bothered with this because my understanding is that a computer can only beat human if the outcomes are limited, and in the game of GO , unlike chess, the outcomes are unlimited so the computer can not calculate the best choice. How does this even work!? Have we really reached AI?

A human mind cannot find the best strategy reliably, so an AI does not need to find the best strategy to win either — by combining machine learning (copying known patterns) with brute force calculation it can make good enough choices to beat most humans.

Designing chips using AI is doable but there is no guarantee it can compete with human designers, at least at the beginning. Over time it is highly likely it will be part of the design process at Apple, Intel etc.