The Performance Inequality Gap, 2024
infrequently.orgVery interesting article.
At first I thought it was going to just be a rant about website bloat and too much JavaScript. I'd have agreed with that, but it wouldn't have taught me anything. But if you keep going, there's much more.
The "Device Performance" section is interesting. I had no idea that iPhones had such an advantage. Now add to that, that the iOS ecosystem is Objective C running native code, whereas the Android ecosystem is Java with JVM overheads! Oof.
(This also explains why I especially hated QR code menus when going out with coworkers. I was using four year old Galaxy A / Moto G class hardware, while my coworkers were using new iPhone Pro class hardware. Their phones were 7.5x faster!)
I now have a medium-high end Android of about a year old. It was a huge step up from my previous phone (the one that struggled at restaurants), but from these charts is nothing special. I will probably keep it for another two years. After that, on the basis of this article, I'll probably shell out the 3x more for a good iPhone. That may be the first time I purchase an Apple product with my own money. (Either that or I'll get a dumb phone and boycott any service that doesn't accept cash. :-))
Even my new Android struggles with some things. Running Firefox, it still struggles on articles from The Guardian for some reason. Everything else is pretty good though.
Smartphones are a funny product whose price goes up as it becomes more essential. Back around 2008, I think, I got a smartphone. I then upgraded every 3-4 years. Initially they were $200. As time went on I started spending $300, $350. The article says that the worldwide average selling price last year was $430. Software is getting slower, and phones are getting more essential, faster than Moore's Law has made them faster or cheaper.
Anyway, the rest of the article has interesting data about bandwidth and more.
Finally, I like the end, where he editorializes a bit.
In conclusion, I guess, if you're a consumer of mobile apps, just get more money and buy a top iPhone (sadly). And if you're a developer, do all your testing on a two year old Moto E, unless you hate poor people.
It's been a long long looooong time that android doesn't use a JVM anymore (but a VM with anyway). It's doesn't even really use "java" at all but kotlin
Moreover, it's a long time that VM have more or less same performances as compiled code (with JIT, AOT, etc.)
But you're free to buy an iPhone if you want. Just use the rights reasons (for example: don't compare a new top of line IPhone with an old budget Android phone... there's high-end android phones too, with performances in line with iphones... and same kind of price tags ;-) )
> there's high-end android phones too, with performances in line with iphones
Yes, but you have to go several years back on the iPhone side: a $300 iPhone 13 from 2021 handily outperforms a new $1k Galaxy S23 Ultra on most web benchmarks and anything else which doesn’t use every core and the GPU heavily.
> there's high-end android phones too, with performances in line with iphones...
TFA shows otherwise, do you have a reason why it would be wrong?
but the article shows top of the line androids being dominated by even old iphones
In Geekbench, which doesn't correspond to real world use. Mid-budget Android phones handily outperformed top of the line iPhones on time to interactive in productivity apps until recently, and now they are merely neck and neck. https://9to5google.com/2023/03/01/galaxy-s23-ultra-speed-tes...
> Smartphones are a funny product whose price goes up as it becomes more essential. Back around 2008, I think, I got a smartphone. I then upgraded every 3-4 years. Initially they were $200
Weren’t those carrier subsidized? You can still get a smartphone for $0 if you’re willing to accept a multi-year contract.
I thought it would be about 10x employees and compensation
Anyone knows what's up the iPhone-Galaxy gap? Is this some bad benchmark? I thought the big Apple revolution was with the M1 in the desktop.
No, it's just a showcase of how absolutely incompetent Qualcomm is at CPU design- they're as bad at CPU design as Intel is at shrinking their node sizes.
It's akin to AMD's inferior CPUs from 2007-2017, except with desktop PCs where you could just buy Intel, Qualcomm is basically the only game in town when it comes to smartphone CPUs to the point where their CPUs are actually more expensive per unit than Apple's (who sell them in 400-dollar SEs both to maintain their edge for businesses, and for people who are looking for a phone that's heads and shoulders a better value than anything in the Android space).
And the thing about Qualcomm is that, because there's no other game in town, they have a captive market and thus don't even need to try (which means the "x86 is doomed ARM is the future" people are just flat wrong in large part because Qualcomm's chips are both inferior to, and more expensive than, what Intel/AMD have on offer).
Sure, they did buy a bunch of ex-Apple chip designers through Nuvia- but they didn't need to because they were going to sell out of their shitty processors anyway- and for Android buyers who are just going to toss the phone after 2 years anyway (and won't ever notice the phone is slow) it's certainly a good strategy.
They aren't incompetent (though I wouldn't call them great either).
They lack vertical integration and are willing to fleece their clients thanks to their position.
Apple's vertical integration means that they can plan chip design in cooperation with final product teams, which means many cost-affecting decisions are made with budget view that includes the whole device and its sales, not predicted markets among chip makers whose total budgets for device designs aren't known.
Apple can easily decide that they are willing to spend a bit extra on, for example, large caches (which have HUGE impact on performance) or very wide superscalar cores, because they are not going to get clients complaining that the chip is now too expensive - and the specific guidance on what design parameters should change is informed from start by the target product.
Meanwhile the closest relation to Apple's chips, Samsung Exynos, for various reasons had its team blocked for cooperation with Samsung Mobile part that designed phones - they effectively had to make most of Exynos SoC in vacuum targeting general markets and hope that selection process doesn't decide Qualcomm's patent and familiarity advantage beats them out resulting in bad financial results which in turn gets management to look badly on them with obvious consequences (I remember how just the fact that they used PowerVR like iPhone at the time led to complaints because games were tested on adreno only).
Neither Qualcomm or Nvidia are providing deep integration at that level (from my understanding, even supposedly "custom" stuff like Qualcomm's S1 and S2 are nowhere near as integrated in design phase as any Apple chip - and Qualcomm seems willing to break Microsoft design for the entire platform with what they delivered). And it was not uncommon to find out way too late that the chip that was selected earlier doesn't really deliver (I recall such complaints specifically about first Snapdragon being massively slower than expected by Google Nexus team, and similar case happening later with Tegra in Xoom. There might have been suggestions by people I talked with that HTC for, possibly informally, banned from consideration by Google Nexus team for a time due to that - Xoom supposedly also made for a bitter pill with nvidia my understanding is that Tegra GPU architecture had some fundamental mismatches with Android)
>because they are not going to get clients complaining that the chip is now too expensive
Or rather, that the only clients who will are end-users and not the phone manufacturers (not that those chips aren't dirt cheap for Apple, since the SE turns a profit). Of course, those phone manufacturers will use them anyway and end up charging some absurd price; it kind of makes sense in that light why Android phones have to be so gimmick-heavy since, just like electric cars, the thing that makes them [barely] work is so ludicrously expensive they have to pack in a bunch of extra stuff to disguise that fact.
>Qualcomm seems willing to break Microsoft design for the entire platform with what they delivered
I guess they didn't really want to compete with Wintel after all. Not that that's a successful business model or anything- why bother with that when you can charge phone equipment manufacturers (which, let's be honest, is the only place those CPUs are ending up these days anyway- the embedded things are all run Android to begin with anyways) absurd prices for bottom-tier performance?
The M1 was not Apple's first custom processor. They have had a performance advantage in mobile for years as the article shows. Especially for single core compute bound workloads.
The latest Snapdragon 8 Gen 3 is not shown in the article but it more or less closed the gap on multi core when compared to A17, but lags behind in single core.
For graphics workloads Android phones are much more competitive and often beat Apple parts.
Where do you think Apple got the expertise and prior experience for hitting a homerun with M1?
The gap was always there, but people don’t tend to look at tech specs on phones the same way as they do on desktops (which makes sense, given that most smartphone users aren’t tech people who would care much bout it).
And specs themselves are often not mapped to real performance on phones too, so it is mostly meaningless. Flagship Androids almost always have more RAM and larger battery capacity than flagship iPhones. But it doesn’t translate into longer battery life or better performance at all (due to power and OS optimization differences, different chip architectures, etc), so those numbers on their own don’t mean much to most people. And even if they cared, it won’t be easy to make a solid comparison.
All I can say from personal experience, as a user I felt that performance gap really acutely when switching from my Galaxy S8+ to iPhone 11.