iOS Optimization w/ CPU vs. Android

[iammike1]

Looking for intelligent discussion here, not a fanboy fest either way.

I've been playing a lot with both my iPhone 4S and my Galaxy Nexus the last month or so. I worry I might wear my SIM out from all the swapping. lol Today I got to thinking about something and was hoping some people with programming experience could help me out here.

iPhone 4S - A5 based on Cortex-A9 clocked at 800MHz / PowerVR SGX543MP2 GPU

Galaxy Nex - TIOMAP based on Cortex-A9 clocked at 1.2GHz / PowerVR SGX540 GPU

I'm not so up to snuff on mobile GPU's but they at least seem like they are from the same series (would appreciate input on this). For the CPU's it appears both are based on the Cortex-A9 dual core but the GNex is clocked at almost double the clock speed stock.

For informational purposes, I've actually rooted my GNex and I'm running franco.Kernel with the CPU overclocked to 1.5GHz and the GPU overclocked to 384MHz (307MHz stock). Not sure on iPhone GPU clock speed.

Here's the meat of my question. I know Apple reaps the benefit of being able to design the hardware and software specifically for each other giving them the ability to virtually eliminate waste. The Galaxy Nexus is about as close as we can get to a Google design with the same two prong approach.

As a result my iPhone 4S feels smoother and faster than my overclocked and optimized GNex on ICS 4.0.4. Is this all there is to it though? Is Apple's advantage in designing both hardware and software hand in hand really so beneficial to overcome 700MHz in user experience?

If you guys can recommend some good cross platform benchmarks that could be used to compare the two beyond just the "feel" of speed, I will gladly perform them and post up the results.

Just looking for some good discussion on how much of a benefit it really is to control the whole potato (Apple) vs Google's approach with Android.

 

[Siggen]

I do think that it is quite a bit difference on the GPU's but I'm not sure. Similar numbers and prefixes suggest a close relation, but even a number bump by one or a new letter suffix can mean a huge difference. Or it can be the same chip remade with smaller more energy sufficient components and with a big improvement in performance in practice (but not theory).

I bet at least one person in here, knows something about the differences of these two compared. just saying.

Apple does not design their own chips, they buy them from samsung, ericsson, and other. Sometimes they pay a lot more to make this model chip unavailable for others, and re-brand it with the apple logo. Usually with a new chip someone cuts it in pieces and looks at it trough a microscope and make an educated guess about a lot of things that is not announced.

The best real life use benchmark would be to install the same game on both and play, take screenshots and compare graphics. And compare framerate feeling. A number on the benchmark is not going to do you any good if the real world application is quite the opposite.

Its not said that just because one is better than the other, the better has the best graphics in game, or the best performance. But real gaming performance is one of the few things that will effect you for real, so it should be taken much more seriously than numbers on a benchmark.

The other thing to remember is take a large sample, not just a few games, take as many games as you can that are able to get that are on both platforms. Even if you don't like the game, try it for the performance test. And write it down in an excel sheet or something with columns something like

|Game Name|Android is better|iPhone is better|Notes|

Make your own columns if you like.



I have worked a few years in the tech industry since before I finished school. And one thing I learned was to chart my clients needs, and present to them the products that match their criteria, and let them chose between them objectively. With great success

 

[w00t951]

1) The Galaxy Nexus' GPU is the same one that was in the iPhone 4, but about 50MHz faster. It's about 7X slower than the iPhone 4S, and Android renders GUI components (in addition to limited hardware acceleration) in a Java environment, which is already eating performance.

2) The Galaxy Nexus' CPU is far faster than the iPhone's A5, but iOS is in C, while Android is in Java. Objective languages are inherently more efficient than languages such as Java.

3) The Nexus is like the iPhone of Android - Google chose and designed the components of the Nexus from scratch. It's irritating that it still has UI lag.

4) Apple designs their A series processors, and Samsung/other company makes them. Get your facts straight.

This link should straighten some things out regarding Android performance.

 

[iammike1]

Thank's w00t! That was a great little read and certainly more educating than the typical "GPU acceleration is everything" argument you hear on the net.

 

[Ray-]

2) The Galaxy Nexus' CPU is far faster than the iPhone's A5, but iOS is in C, while Android is in Java. Objective languages are inherently more efficient than languages such as Java.

Well that's not entirely true. Android is built on Linux, which means most of the OS itself is in C or C++. It's just that google required the apps that run on it to be coded with a variation of Java. iOS on the other hand is based on Darwin (which itself is from NEXTStep, from when Jobs was at NEXt in the 80s!), which is a Unix OS, so also in C and C++.

I used a Nexus One for almost 2 years before switching to the 4S, and from a user perspective, the 2 OS's are really more similar than they are different.

I'm sure none of us has looked at both the code inside iOS and Android (a little hard since iOS is closed source). Now I might sound like a bigot, but if you asked me for a personal opinion why iOS feels snappier than Android...Google probably needed to clone iOS, but they couldn't exactly clone how everything was done verbatim, so they had to look for alternatives, which might not be optimal.

 

[Technarchy]

Android is not optimized for anything.

It wastes resources, suffers from massive memory leaks, and requires twice the hardware of iOS, or WP7 to provide a semi acceptable user experience.

All around, it's just a sloppy endeavor.

 

[w00t951]

Well that's not entirely true. Android is built on Linux, which means most of the OS itself is in C or C++. It's just that google required the apps that run on it to be coded with a variation of Java. iOS on the other hand is based on Darwin (which itself is from NEXTStep, from when Jobs was at NEXt in the 80s!), which is a Unix OS, so also in C and C++.

Java is a versatile platform that can be run on multiple devices without much modification - but when the phone's apps and GUI are both Java-based, you end up with an inconsistent user experience and huge hardware requirements.

 

[Siggen]

4) Apple designs their A series processors, and Samsung/other company makes them. Get your facts straight.

Common! Are you trying to get away with Apple designing the processor? Nope nope nope. Its ARM and Samsung. If by designs you mean sits in multiple meetings with Samsung and ARM, and also sits in multiple meetings internally to decide if they wish to continue with Samsung again or chose another vendor.

Its not as simple as setting up its own department to design components on a 16-65 nanometer scale.

Java is a versatile platform that can be run on multiple devices without much modification - but when the phone's apps and GUI are both Java-based, you end up with an inconsistent user experience and huge hardware requirements.

You are exactly right here. The way java executes codes is similar to a emulating a virtual machine which runs the code. Scratch the part about similar, that's exactly how it works..

 

[w00t951]

Common! Are you trying to get away with Apple designing the processor? Nope nope nope. Its ARM and Samsung. If by designs you mean sits in multiple meetings with Samsung and ARM, and also sits in multiple meetings internally to decide if they wish to continue with Samsung again or chose another vendor.

I suppose Apple was lying in 2010 when they said that the A4 was a completely Apple-designed chip? Sure, it was manufactured by Samsung, but that's because Apple didn't have a chip manufacturing line (now they do).

I doubt the A5 is any different, and according to Wikipedia, it's not. Still designed by Apple, Inc.

I'm just saying, but Apple 1) designs the A4 2) designed the A5 and 3) bought a chip manufacturing company specifically to design the A4.

 

[Intell]

Isn't the Galaxy Nexus' GPU a single core variation of the one that's used in the A5?

 

[w00t951]

Isn't the Galaxy Nexus' GPU a single core variation of the one that's used in the A5?

No, it's an SGX540. The iPhone uses something like the SGX544MP2. I'm too lazy to find the actual specs, but it's most definitely a completely different chip.

 

[jamesrt2004]

I suppose Apple was lying in 2010 when they said that the A4 was a completely Apple-designed chip? Sure, it was manufactured by Samsung, but that's because Apple didn't have a chip manufacturing line (now they do).

I doubt the A5 is any different, and according to Wikipedia, it's not. Still designed by Apple, Inc.

I'm just saying, but Apple 1) designs the A4 2) designed the A5 and 3) bought a chip manufacturing company specifically to design the A4.

it is mostly designed by apple correct. They have a lot of great engineers that work in this sort of thing.

They will be liaising and working very very close with Samsung on the chips to get the most efficient chip/build-able chip for the $$$

You could say like in football(soccer) if someone upfront shoots and it takes a small deflection they will still claim the goal..

samsung may guide apple to a better chip but ultimately it's apples design.

 

[cynics]

I wouldn't even bother with benchmark programs that can run on the phone.

Before:

After an app update using the same input criteria.

With multithread turned on just incase your wondering

Point is if there are obvious differences from one update to the next on the same OS how accurate can it be cross platform?

 

[ChazUK]

I'd guess GLBenchmark would be worth a try if you wanted some cross platform benchmarks:

https://play.google.com/store/apps/details?id=com.glbenchmark.glbenchmark21&hl=en

http://itunes.apple.com/gb/app/glbenchmark-2.1/id460948646?mt=8

The Galaxy Nexus is pushing around an awful lot of pixels for a relatively weak GPU which might explain some of the lagging you're experiencing. I've done three tests with Nenamark 2 and GLBenchmark 2.1 and get the following:

Nenamark:
Nexus S: 24.4fps
Galaxy Nexus: 24.5fps
Galaxy S III: 58.8fps (VSync limited)

GLBenchmark (high)
Nexus S: 2843 frames, 25fps
Galaxy Nexus: 2256 frames, 20fps
Galaxy S III: 6479 frames, 57fps (limited by Vsync)

GLBenchmark (offscreen)
Nexus S: 2064 frames, 18fps
Galaxy Nexus: 3154 frames, 25fps
Galaxy S III: 11190 frames, 99fps

It seems to me that the Galaxy Nexus GPU higher clocked SGX 540 at 1184x720 resolution is performing as good as the Nexus S's 200MHz SGX450 running at 800x480.

 

[nickchallis92]

i wish the S3 looked as good on the outside as it does on the inside

 

[kdarling]

Here's the meat of my question. I know Apple reaps the benefit of being able to design the hardware and software specifically for each other giving them the ability to virtually eliminate waste.

Depends on how much Apple writes the code with an eye towards the future. One whole point of an OS is give yourself some hardware independence.

As for Android manufacturers, they have the source code, so they're free to also optimize sections of the OS for their own hardware, and some no doubt do so. Along the way, they sometimes even add their own optimized features, such as Samsung did with their play-anywhere video window.

As a result my iPhone 4S feels smoother and faster than my overclocked and optimized GNex on ICS 4.0.4. Is this all there is to it though?

I think a lot of people expect machines to be instantaneous all the time. I'm not bothered by occasional lag (which I've gotten on iOS as well), because I realize that sometimes I'm asking the phone to do multiple things at once, whether it be background downloading or constant locating or whatever.

 

[DodgeV83]

I wouldn't even bother with benchmark programs that can run on the phone.

Before:

Image

After an app update using the same input criteria.

Image

With multithread turned on just incase your wondering

Image

Point is if there are obvious differences from one update to the next on the same OS how accurate can it be cross platform?

We talked about this already

There is a known issue in the Linpack benchmarks, on all smartphone platforms, that renders the results inaccurate if the Problem Size is set too low. Something about the processors being too fast to get a good reading from such a small Problem Size. The Android app uses a Problem Size of about 550.

Use a Problem Size of 550 in your app, and you'll see the numbers change. Anything above 550 gives a consistent score.

Wither the benchmark itself is worth looking at is up for debate, as lilo777 pointed out iOS has a built-in function for the Linpack calculations, where Android devs have to build it manually, but it should not be discounted for the reason stated.

 

[DodgeV83]

If you guys can recommend some good cross platform benchmarks that could be used to compare the two beyond just the "feel" of speed, I will gladly perform them and post up the results.

Just looking for some good discussion on how much of a benefit it really is to control the whole potato (Apple) vs Google's approach with Android.

The best GPU benchmark is GLBenchmark, but don't buy the app, just check their Results page:

http://glbenchmark.com/result.jsp

For CPU benchmarks, the only one I am aware of on both platforms, is Linpack. That benchmark has been debated on these forums a few times, with some saying its not relevant, and some saying it is.

If iOS has a built-in super efficient function for Linpack calculations, and Android does not, this can give an unfair advantage to iOS if you're trying to compare CPUs. That doesn't seem to be your comparison here, however, and that's the million dollar question.

If iOS is tuned so specifically to its hardware, that the OS and Apps are able to have better performance on older hardware, which one is "faster"?

 

[naths]

when google released the galaxy nexus it was not particularly well received, as in the GS2 was better phone and almost a year older, not a very high spec phone for a nexus, its just a developers phone for Android really, no overlay of touch-wiz or sense, just ICS straight from google, if you were to do the same test with the GS3 or the HTC one X i think the 4S would be struggling...

 

[iammike1]

Thanks for all the constructive input everyone. I was worried this would just turn into a bias jerk fest but instead, good conversation is being had.

I didn't realize that the iOS GPU was such a beast compared to the GNex GPU which sounds like it's just a slightly upclocked version of what was in the iPhone 4. Considering that, I would say Google actually did a damn good job keeping things smooth 90% of the time.

Of course it doesn't hurt that I'm running the CPU and GPU overclocked. lol

I'm getting ready for work right now but after work, I might run a few benchmarks just for ***** and giggles to compared the OC'd GNex to the iPhone 4S.

I'm pretty much back on my iPhone full time now though. There are things I really do like about ICS (I hated GB so bad) but it's the app quality that keeps bringing me back to iOS.

----------

when google released the galaxy nexus it was not particularly well received, as in the GS2 was better phone and almost a year older, not a very high spec phone for a nexus, its just a developers phone for Android really, no overlay of touch-wiz or sense, just ICS straight from google, if you were to do the same test with the GS3 or the HTC one X i think the 4S would be struggling...

True, although I certainly wouldn't consider keeping touchwiz or sense off the phone a negative.

 

[naths]

yep...true, touch-wiz is a right pain, but HTC have done great with sense 4.0...iv got a One X and a GS3 both on ICS and the One X is a better experience.....

 

[gladoscc]

Download Geekbench.

Also, Android is less optimized than iOS because it has to run on more hardware.

 

[lilo777]

Download Geekbench.

Also, Android is less optimized than iOS because it has to run on more hardware.

Samsung version of Android runs only on Samsung phones. Not so different from Apple. Regardless, as I understand, Galaxy SIII beats iPhone 4S in all benchmarks and in some cases by a huge margin.

As to the original question - one thing that was not mentioned here is the memory subsystem in the CPU. Performance depends not just on CPU frequency. Sure, these CPUs have similar architectures but they are not identical. Apple chips have faster memory interface and this helps a lot for certain tasks.

 

[blackhand1001]

1) The Galaxy Nexus' GPU is the same one that was in the iPhone 4, but about 50MHz faster. It's about 7X slower than the iPhone 4S, and Android renders GUI components (in addition to limited hardware acceleration) in a Java environment, which is already eating performance.

2) The Galaxy Nexus' CPU is far faster than the iPhone's A5, but iOS is in C, while Android is in Java. Objective languages are inherently more efficient than languages such as Java.

3) The Nexus is like the iPhone of Android - Google chose and designed the components of the Nexus from scratch. It's irritating that it still has UI lag.

4) Apple designs their A series processors, and Samsung/other company makes them. Get your facts straight.

This link should straighten some things out regarding Android performance.

The iPhone 4's gpu is not the same as the nexus's. The iPhone 4 uses a power sgx 535 which is a much slower than the powersgx 540 even when both clocked at 200mhz which is what the nexus S's (sgx540) and the iPhone 4's (sgx535)are clocked at. The Galaxy nexus's is clocked at 300mhz. The difference between the powersgx 535 and powersgx 540 at the same clock speed roughly 200%. The nexus is clocked higher as well which increases its performance more.

 

[iammike1]

Samsung version of Android runs only on Samsung phones. Not so different from Apple. Regardless, as I understand, Galaxy SIII beats iPhone 4S in all benchmarks and in some cases by a huge margin.

As to the original question - one thing that was not mentioned here is the memory subsystem in the CPU. Performance depends not just on CPU frequency. Sure, these CPUs have similar architectures but they are not identical. Apple chips have faster memory interface and this helps a lot for certain tasks.

It's not that Samsung's version of Android runs only on Samsung phones, it's that the layer of Samsung crapwiz they throw on top of Android runs only on Samsung phones and even that is debatable. Many parts of Touchwiz will work on non Samsung phones when stripped out of the overall Touchwiz picture.

Also, I would hope like hell the SIII could beat the iPhone in benchmarks. Quad-core 1.4GHz Xenos processor, some nasty good GPU from what I've heard, and it's like 8 months younger than the iPhone 4S. What's disappointing to me and what led to the kinds of questions I've been talking about in this thread is that it took that much crazy mobile power to start out-doing the 4S.

An unfortunate side effect of this is power consumption. My Galaxy Nexus battery life isn't nearly as bad as a lot of Android haters would have had me believe before I tried for myself...but it is reasonably worse than my 4S while having around 300mAh more juice in the standard battery...and the GNex is no power house when compared to the likes of the HOX and the SGSIII.

Makes me wonder what could be achieved if they could get even close to the efficiency of iOS while still packing those processors and GPU's in the Android phones. I think it would be freaking amazing.

----------

...there is a part of me that would love to go out today and buy an international version SIII here in Japan but really, my biggest hold up is the screen size.

iPhone needs a bigger screen, no doubt...but the heights of screen size Android flagship phones are getting to are just dumb.

Yeah, they are super pretty and great for watching video...but to actually USE the device in day to day situations presents more than just difficulty in reaching parts of the screen. I've almost dropped my GNex several times because I was stretching my hand to reach a button on the top of the screen while only having one hand available. I didn't like that feeling and the GNex is a 4.65 screen. I'm scared to think what would happen with a 4.8 inch screen.

Although I've yet to get to try one myself, I think 4.3 to 4.5 would be a perfect size for me personally. For others this would vary of course. I just wanna know when the only difference between an Android phone and an Android tablet will be the ability to make a phone call.

----------

The iPhone 4's gpu is not the same as the nexus's. The iPhone 4 uses a power sgx 535 which is a much slower than the powersgx 540 even when both clocked at 200mhz which is what the nexus S's (sgx540) and the iPhone 4's (sgx535)are clocked at. The Galaxy nexus's is clocked at 300mhz. The difference between the powersgx 535 and powersgx 540 at the same clock speed roughly 200%. The nexus is clocked higher as well which increases its performance more.

Then you toss a quality kernel in the picture and get even more out of it.

My GNex GPU is running at 387MHz right now and according to those devs who are smarter than me, they were able to do it without changing the power profiles from the 307MHz stock and have avoided an impact on battery while increasing performance.

I won't lie, the Android root devs give me joy sometimes.