Raw Galaxy S4 Vs HTC One Performance discussion

[Kevin OQuinn]

So you're suggesting that I secretly handicap both devices by eliminating optimizations completed by both HTC and Samsung, not to mention put the gs4 on an os for which oft was never intended? And not to disparage Rom makers of the world...but I think you'd have a bigger argument on your hands about "objectivity" and potential cheats in those ROMs...not because of anyone doing anything below board..but because you'd be looking at pieces of work largely unqualified buy anyone but that developers. If anything, I think that looking at the nexus editions of those are made available...at least it's be a level(er) playing field.

Posted via Android Central App

You yourself said you wanted to talk about the metal and figure out what the difference in performance could be contributed to. Then you refuse to eliminate software as a potential variable. Tell me again what you're trying to accomplish?

Yes, the Google Editions would be great for this type of comparison, but really, the S4 has a higher clock speed. Did anyone expect it to score lower on benchmarks? When those come out, and if the difference in scores between the two are smaller, can we lay this to rest with the fact that the difference is software?

Also, it is well known that nVidia and ATI optimized for specific synthetic benchmarks. Obviously it's really easy to get caught, since they did.

Sent from my HTC One using Tapatalk 4 Beta

 

[slackerjack]

You yourself said you wanted to talk about the metal and figure out what the difference in performance could be contributed to. Then you refuse to eliminate software as a potential variable. Tell me again what you're trying to accomplish?

Yes, the Google Editions would be great for this type of comparison, but really, the S4 has a higher clock speed. Did anyone expect it to score lower on benchmarks? When those come out, and if the difference in scores between the two are smaller, can we lay this to rest with the fact that the difference is software?

Also, it is well known that nVidia and ATI optimized for specific synthetic benchmarks. Obviously it's really easy to get caught, since they did.

Sent from my HTC One using Tapatalk 4 Beta

Putting them on equal platforms only works if its not some half baked homebrew OS, again...not a knock at the developers...but reliable as a test bed? Just no. Both HTC and Samsung will have some low level driver control even in AOSP, that does make a fair(er) test.

And yes, I'd certainly consider putting the s4 on 4.1 for which drivers were never written by the manufacturer and have to get hacked to get functional a legitimate disadvantage, so I agree that the developer editions are going to be the best level testbed. You are dreaming, however if you think the performance gap is going to do anything but widen under AOSP, as you mentioned...clock speed is a differentiator, but doesn't explain a much larger difference in performance than most of you are willing to admit in flat testing...but the fact that its NOT like for like hardware, a fact that everyone outside of Anandtech just glazes over and puts in the "it shouldn't matter that much" category will likely rear its head again. Here's hoping I get to eat my words...I actually like to be proven wrong. But I'm rarely offmy game as it pertains to this subject, in fact...people pay me for my expertise in that arena.

For the record, I hold a PhD in mathematics, a BFA in CGI animation...I worked for both ATI and Nvidia in the last 6 years and was with 3dfx before that. My primary specialty is AF and adaptive AA algorithms in driver support. You won't find many people who are as a qualified nerd as me! Who knew a well thought out plan to play videogames for the rest of my life would turn into actual work.

Jack

Posted via Android Central App

 

[droric]

I question how you know the sensors are still being used? I know i tested the 'Air view' sensor by turning on the 'Quick glance' feature and checking the IR emitter with a camera phone (see's IR as visible light). Then I turned off the 'quick glance' setting and the IR light ceased to be activated while the phone was off. Are you sure there features you missed that are keeping the sensors on?

The problem with the S4 might be the sensors. Even when turned off, they apparently draw the power, which means they are probably polling, but just not giving user output.

Just a thought, since the other issue could be the gpu handling the data to a1080 display, but the One apparently does not have the issues.

 

[natehoy]

Putting them on equal platforms only works if its not some half baked homebrew OS, again...not a knock at the developers...but reliable as a test bed? Just no. Both HTC and Samsung will have some low level driver control even in AOSP, that does make a fair(er) test.

And yes, I'd certainly consider putting the s4 on 4.1 for which drivers were never written by the manufacturer and have to get hacked to get functional a legitimate disadvantage, so I agree that the developer editions are going to be the best level testbed. You are dreaming, however if you think the performance gap is going to do anything but widen under AOSP, as you mentioned...clock speed is a differentiator, but doesn't explain a much larger difference in performance than most of you are willing to admit in flat testing...but the fact that its NOT like for like hardware, a fact that everyone outside of Anandtech just glazes over and puts in the "it shouldn't matter that much" category will likely rear its head again. Here's hoping I get to eat my words...I actually like to be proven wrong. But I'm rarely offmy game as it pertains to this subject, in fact...people pay me for my expertise in that arena.

For the record, I hold a PhD in mathematics, a BFA in CGI animation...I worked for both ATI and Nvidia in the last 6 years and was with 3dfx before that. My primary specialty is AF and adaptive AA algorithms in driver support. You won't find many people who are as a qualified nerd as me! Who knew a well thought out plan to play videogames for the rest of my life would turn into actual work.

Jack

Posted via Android Central App

For someone with so many letters after your name, I don't get what you are trying to compare, then. You are testing two different devices with a known delta in hardware specs between them and asking why the actual delta in performance is so much higher.

The only factor is either other components such as faster memory or an optimized architecture between the chipset and other components, or differences in the hardware drivers. Given that the two are running different Android kernels and were built over the same chipset but at different clockspeeds at different factories with different design parameters, you'll have to get detailed hardware specs for each device to know for sure.

But, honestly, given that the "unexpectedly faster" machine is running a newer kernel and newer version of the OS overlay, and ALSO given that patches have come out for the "unexpectedly faster" machine that fix some of the lagginess (which you'd expect might increase performance), I think it's very reasonable to assume that software can account for the performance difference.

Driver support is very important, as I'd hope you would know, and 4.1 and 4.2 would by necessity have different drivers for that same chipset.

 

[slackerjack]

Tell you what, you show me some performance numbers that illustrated that pure AOSP 4.2.1 is faster than 4.1.2 and by very consistent 15% across the board...I'll eat my words.

Lol, I'm pretty sure if that were true, Google would be talking about that and in public, a lot..think of the marketing..."Project butter just got 15% butterier". I sure as hell would be screaming it from the rooftops!

Posted via Android Central App

 

[android010]

I will give my insight on this issue. I have been following the tests videos for a long time. They are fun to watch but in the real life they mean - to me- almost nothing. I came up with this conclusion after switching from the nexus 4 to the galaxy s4 . S4 beats nexus 4 in every test by a big margin. You would think that you will have a much smoother experience on the S4.

However, after switching I came to experience a LOT of lag and crashes here and there. Not a single day can go without at least three or four crashes and lags. This pissed me off a big deal coming especially coming from the Nexus 4 which I never experienced any lag with.

This might be the case with my phone only but it is true and sadly painful to spend a lot of money on such a high end phone to end up with stutter and lag and crashes. I hope that samsung figures this out with a software update or something or I might seriously consider going back to the nexus line.

 

[slackerjack]

Sounds like you have a bum device. I haven't experienced any of the crashes that you're describing, consider getting it exchanged.

Posted via Android Central App

 

[snookasnoo]

Something real world and repeatable. Most of these mean very little considering you have Sense on one device and Touchwiz on another.

 

[mrbootcrm]

Just glanced through and am wondering why it matters? Why performance numbers would ever take priority over user experience is beyond me. Do people need raw computing power on their phones for some purpose I'm unaware of?

 

[Jerry Hildenbrand]

If you're truly interested in testing metal, and not just trying to make your shiny new S4 look good, load the exact same ROM on each phone and run your tests again. Until then you can't rule out any cheating from either phone/software (not that I believe either is).

But you're still running benchmark apps through an open source Java VM.

You can't benchmark Android devices using anything that has a gui. Doing so is the same as running two different versions of the benchmark program.

But I have a solution for the OP, if he's actually serious that is.

Compile the "time" binary for ARM to run on a generic Linux 3.0 install. Any Linux guru or Android developer can help.
Use the same kernel version on each device. Any kernel dev can help.
Shell into your phone, and use the time binary to see how long the system takes to perform a task that doesn't need the ui. You can do this part yourself.

./adb shell
su
cd /
/usr/xbin/time cd/system

The response will be three lines with:

• real time
• user time
• system time

in seconds. Smaller value means it performed things faster.

 

[natehoy]

Just glanced through and am wondering why it matters? Why performance numbers would ever take priority over user experience is beyond me. Do people need raw computing power on their phones for some purpose I'm unaware of?

Honestly, me, too. But it is interesting to see a performance difference above and beyond what you'd expect from the hardware specs. The trouble is, that difference can easily be explained by the fact that the two platforms are running entirely different software.

 

[planoman]

I will give my insight on this issue. I have been following the tests videos for a long time. They are fun to watch but in the real life they mean - to me- almost nothing. I came up with this conclusion after switching from the nexus 4 to the galaxy s4 . S4 beats nexus 4 in every test by a big margin. You would think that you will have a much smoother experience on the S4.

However, after switching I came to experience a LOT of lag and crashes here and there. Not a single day can go without at least three or four crashes and lags. This pissed me off a big deal coming especially coming from the Nexus 4 which I never experienced any lag with.

This might be the case with my phone only but it is true and sadly painful to spend a lot of money on such a high end phone to end up with stutter and lag and crashes. I hope that samsung figures this out with a software update or something or I might seriously consider going back to the nexus line.

The nexus 4 is very smooth, but I might suggest that something you may have done to your s4 is making it crash. I have heard of lag but not repeated crashes. Something is wrong with your phone.

I have not had any desire to touch my nexus 4 since getting my s4 on May 13. It is that good.

Sent from my Nexus 7

 

[slackerjack]

But you're still running benchmark apps through an open source Java VM.

You can't benchmark Android devices using anything that has a gui. Doing so is the same as running two different versions of the benchmark program.

But I have a solution for the OP, if he's actually serious that is.

Compile the "time" binary for ARM to run on a generic Linux 3.0 install. Any Linux guru or Android developer can help.
Use the same kernel version on each device. Any kernel dev can help.
Shell into your phone, and use the time binary to see how long the system takes to perform a task that doesn't need the ui. You can do this part yourself.

./adb shell
su
cd /
/usr/xbin/time cd/system

The response will be three lines with:

• real time
• user time
• system time

in seconds. Smaller value means it performed things faster.

Thanks for the idea, I'll get cracking on it this weekend, have a lot going on at present bit should be able to get something built over the next week or so!

Posted via Android Central App

 

[Johnly]

But you're still running benchmark apps through an open source Java VM.

You can't benchmark Android devices using anything that has a gui. Doing so is the same as running two different versions of the benchmark program.

But I have a solution for the OP, if he's actually serious that is.

Compile the "time" binary for ARM to run on a generic Linux 3.0 install. Any Linux guru or Android developer can help.
Use the same kernel version on each device. Any kernel dev can help.
Shell into your phone, and use the time binary to see how long the system takes to perform a task that doesn't need the ui. You can do this part yourself.

./adb shell
su
cd /
/usr/xbin/time cd/system

The response will be three lines with:

• real time
• user time
• system time

in seconds. Smaller value means it performed things faster.

Now this is smart. It is leveling software to "truly" test the hardware. The OS will be the same in a sense <~~~~metaphorical. Thanks Jerry for injecting caffine into this thread
Sent from my SCH-I605 using Tapatalk 2