Comment by simonh | Hacker News Reader

Comment by simonh | Hacker News Reader

Parent Full thread

Performance and features. In single threaded performance, Apple's A chips are way ahead. That's because Apple highly customizes and optimizes their core designs, while Qualcom and Samsung use largely vanila ARM reference designs with maybe a few tweaks. Apple chips also have considerably more cache. That's difficult for the competition to counter because they are very price sensitive and cache size eats up expensive die space.

The counter argument is that the other systems make up for this by having more cores, which is really a cop out. Single threaded execution has a far more direct effect on user experience, gaming performance, etc. They can't compete in core engineering so make it up by slapping on more fairly generic cores. It does appear that Samsung is responding to this and investing in more advanced core designs.

The final area is custom features like the neural engine behind real-time face recognition, real time 3D face lighting effects and such. The competition don't really have these at all, so we don't know how far behind they are.

There's a pretty good article on this linked below.

https://seekingalpha.com/article/4138071-apple-cpu-advantage...

>That's because Apple highly customizes and optimizes their core designs, while Qualcom and Samsung use largely vanila ARM reference designs with maybe a few tweaks

Here's the Anandtech article on the "few" tweaks Samsung has made to its 6 wide decode custom M3 cores in the upcoming Exynos 9810.

https://www.anandtech.com/show/12361/samsung-exynos-m3-archi...

The Samsung M1, M2 and soon the M3 are all fully custom.

Yep, it does look like Samsung may be in a position to catch up soon. That could make a real problem for Qualcom the other Android manufacturers. Samsung are the biggest manufacturer of high end Android phones. If Qualcom can't sell high end CPUs to them any more, it might make manufacturing any truly high end CPUs uneconomical, or at least drive up costs even further.

Watch out for what comes out of ARM Austin. Their last core was the A72 from a couple of years ago. Rumour has it they've been working on a big core to better compete with Apple and Samsung.

Looking back, it makes sense that Qualcomm discarded their underperforming custom design efforts (at least for the mobile market). If ARM can deliver a competitive design, why not fully commit to their roadmap and save significant amounts on R&D costs?

Because if ARM cannot deliver they are fucked.

A lesson Apple learned at the hands of Motorola.

I sometimes wonder how much of It Just Works was influenced by having slower machines. When doing everything takes longer, if you do it right the first time then it’s still faster than doing it twice.

ARM is not just one company - it's a semi-standard.

Would you mind backing up your claim? To my knowledge, ARM is one company that licenses their designs to other companies, who are free to parameterize and modify as they please.

I don't know what the above comment was referring to but they are more than what you describe. I'd submit AMBA as an example of this; I've seen this tech utilised by IP for ASICs that don't actually contain any ARM processor cores themselves.

https://en.m.wikipedia.org/wiki/Advanced_Microcontroller_Bus...

I think the idea is that it probably matters very little to Apple or Samsung or any other company that has an architectural license how well the ARM originated core performs because they are building their own core and ARM is basically a standard committee to them. Obviously companies that are using the actual ARM cores care about ARM's ability to perform.

Why don't companies just come up with their own designs?

First off designing a whole core architecture from scratch is a huge and very expensive undertaking. Youd need to be able to get a really big competitive advantage to make it worthwhile.

Secondly theres a large ecosystem of add on components designed work with ARM and be dropped into ARM SOCS, such as GPUs, Wifi modules, Gyroscopes, wireless modems, GPS, etc.

Thirdly there are a lot of SOC engineers very familiar with ARM. You can hire them streight from competitors, or college, including PHDs that have done research on it. Youd need to train up any new hirs from scratch on your architecture.

Finally theres a huge software development tool chain built around the established processor architectures. To support a new architecture youd also need to build a set of compiler back ends, bearing in mind the existing ones benefit from many years of tweaking and optimization.

Yes, SW is the key; especially Linux. ARM is way ahead of other embedded CPU architectures in terms of Linux support.

There are two types of ARM licenses. One is an Architectectural license. This allows licensees to design and implement the ARM ISA. Apple and QCOM among others have Architectural licenses. They can add customizations such as # pipelines stages. This license is very expensive.

The other license is getting ARM's implementation of a CPU (ie A57, A72 etc). Customers can parametize # cores, cache size, memory width etc, but the basic architecture is fixed.

Because it gives you no competitive advantage.

>Performance and features.

I think both these metrics are now adequate, battery life is the single metric I am interested in now.

Battery life is, realistically, part of performance- it’s just “per watt”. Apple is also a leader in that area as well.

Apple made the decision to pursue performance at the cost of device life.

Reply on news.ycombinator.com