Root For all those wondering about MFLOPS/OC

[reinert012]

*Please not this is only my phone and my opinion, results may vary

I have ran a few tests with Linpack for Android, free on the marketplace.

My goal of this is for people to see exactly what average speeds my phone is running at given the CPU speed, and hopefully they will have an idea of how fast their phone will be with the same CPU speed and it will clear up some questions on exactly what a good MFLOP speed is.

I am running Aloysius Snow found here:
[ROM] Aloysius 2.0 Snow with OC, Friendstream and APP2SD (6/2/10) - xda-developers

Thanks to a great dude i have JIT enabled and the updated kernal, tutorial found here:
http://androidforums.com/all-things-root-eris/90874-how-remove-senseui-enable-jit.html#post859986

I ran 10 tests each at:
528 (Eris default)
710
729
748
768
806

Here are my averaged results
528-3.4566 MFLOPS
710-4.7158 MFLOPS
729-4.8542 MFLOPS
748-4.9942 MFLOPS
768-5.1346 MFLOPS
806-5.2922 MFLOPS


Conclusion (this is where my opinion comes in)
It is obvious that running the Eris's default 528 is going to have its limitations. But as i went from 710 to 806 it was only .5764 MFLOPS in difference, and to be honest i cannot see a big difference at all from 710 to 806. Im sure running at a higher CPU speed will shine in some areas with heavy CPU usage, but for me, it is definitely not worth the battery drain. My battery drains a lot quicker if i am running at 748+ when the phone is on and being used, not idle and not in standby. I think I will be moving back down to 710 or 729.


If you are not running the updated kernal and do not have JIT enabled like in the link above, your phone will have significantly less numbers being tested at same CPU speeds as mine listed.

**DISCLAIMER** some phones might crash past the 710 CPU speed, I am not responsible if your phone bricks, have a nice day!

 

[scary alien]

Very nice work. Thanks for taking the time to research and post. I'm still running Ivan's Eris Offical 1.0 w/OC at its standard 710 MHz and have been very happy with its performance and battery life. My linpack MFLOPS is right around 3.0 on average. Haven't (yet) seen a need to crank it up higher or pursue a JIT ROM.

 

[ACD168]

I tend to agree with this I never overclock over 710, but that's just me.

 

[reinert012]

Very nice work. Thanks for taking the time to research and post. I'm still running Ivan's Eris Offical 1.0 w/OC at its standard 710 MHz and have been very happy with its performance and battery life. My linpack MFLOPS is right around 3.0 on average. Haven't (yet) seen a need to crank it up higher or pursue a JIT ROM.

I used to run Ivans, it was actually the first ROM i ran and i loved it. But i loved the look of Aloyrius, and was bored one day and did that JIT update haha

 

[scary alien]

I used to run Ivans, it was actually the first ROM i ran and i loved it. But i loved the look of Aloyrius, and was bored one day and did that JIT update haha

I'm hoping that Ivanmmj will release an updated version (2.2 maybe ...? ) [shameless request <grin>] that will have JIT and some other great goodies. I know that JIT is still "under construction" and doesn't play well with Sense, so I'm happy to wait until most of the kinks are worked-out.

My main goal is a rock-solid phone / ROM which I have now. When / if I ever upgrade to "the-next-big-thing" I'll feel like I have the freedom to tweak-out my Eris and really play with some custom ROMs (and hopefully Froyo).

 

[erisuser1]

Here are my averaged results
528-3.4566 MFLOPS
710-4.7158 MFLOPS
729-4.8542 MFLOPS
748-4.9942 MFLOPS
768-5.1346 MFLOPS
806-5.2922 MFLOPS

I notice that 806 / 528 = 1.53 and that 5.2922 / 3.4566 = 1.53.

That's sort of exactly what you would expect - a 53% improvement over that range of speeds, assuming that the CPU is constantly clocked at a fixed maximum speed.

OTOH, your gut feel about power dissipation increasing faster than the gain in processing speed is probably quite correct - as the clocks are hiked up, the logic supply voltage is also increased (in steps) at various clock speeds - and the power dissipation goes as V^2 (per clock). If the supply voltage was fixed, this wouldn't matter - but since it is intentionally increased at higher clock speeds (to preserve signal margins) - you actually get less computational work done per unit of battery reserve at higher clock speeds.

Or, to put it in simpler terms, if you were getting 12 hours of use running at 528, and you could increase the clock speed by 53%, then you won't necessarily get 8 hours of use - you might only get 5 or 6 hours. Yes, the battery is going to drain quickly.

Actually, what seems more interesting is that without JIT, most of the Eris ROMs come in with a MFLOPs figure around 3 @ 710 Mhz - note that the JIT-enabled version of the Dalvik engine shows a whopping 4.7 / 3 = 1.56 increase over non-JIT machines.


Personally, I suspect that the JIT vs. non-JIT result is not much more than a pathology of that particular benchmark, and little more than that. I suppose I'd love to be proved wrong about that... because it would means that our little Eris keeps getting better and better with new software!

eu1

 

[reinert012]

I notice that 806 / 528 = 1.53 and that 5.2922 / 3.4566 = 1.53.

That's sort of exactly what you would expect - a 53% improvement over that range of speeds, assuming that the CPU is constantly clocked at a fixed maximum speed.

OTOH, your gut feel about power dissipation increasing faster than the gain in processing speed is probably quite correct - as the clocks are hiked up, the logic supply voltage is also increased (in steps) at various clock speeds - and the power dissipation goes as V^2 (per clock). If the supply voltage was fixed, this wouldn't matter - but since it is intentionally increased at higher clock speeds (to preserve signal margins) - you actually get less computational work done per unit of battery reserve at higher clock speeds.

Or, to put it in simpler terms, if you were getting 12 hours of use running at 528, and you could increase the clock speed by 53%, then you won't necessarily get 8 hours of use - you might only get 5 or 6 hours. Yes, the battery is going to drain quickly.

Actually, what seems more interesting is that without JIT, most of the Eris ROMs come in with a MFLOPs figure around 3 @ 710 Mhz - note that the JIT-enabled version of the Dalvik engine shows a whopping 4.7 / 3 = 1.56 increase over non-JIT machines.


Personally, I suspect that the JIT vs. non-JIT result is not much more than a pathology of that particular benchmark, and little more than that. I suppose I'd love to be proved wrong about that... because it would means that our little Eris keeps getting better and better with new software!

eu1

Wow i did not notice that, that is pretty crazy.

 

[andrizoid]

*Please not this is only my phone and my opinion, results may vary

I have ran a few tests with Linpack for Android, free on the marketplace.

My goal of this is for people to see exactly what average speeds my phone is running at given the CPU speed, and hopefully they will have an idea of how fast their phone will be with the same CPU speed and it will clear up some questions on exactly what a good MFLOP speed is.

I am running Aloysius Snow found here:
[ROM] Aloysius 2.0 Snow with OC, Friendstream and APP2SD (6/2/10) - xda-developers

Thanks to a great dude i have JIT enabled and the updated kernal, tutorial found here:
http://androidforums.com/all-things-root-eris/90874-how-remove-senseui-enable-jit.html#post859986

I ran 10 tests each at:
528 (Eris default)
710
729
748
768
806

Here are my averaged results
528-3.4566 MFLOPS
710-4.7158 MFLOPS
729-4.8542 MFLOPS
748-4.9942 MFLOPS
768-5.1346 MFLOPS
806-5.2922 MFLOPS


Conclusion (this is where my opinion comes in)
It is obvious that running the Eris's default 528 is going to have its limitations. But as i went from 710 to 806 it was only .5764 MFLOPS in difference, and to be honest i cannot see a big difference at all from 710 to 806. Im sure running at a higher CPU speed will shine in some areas with heavy CPU usage, but for me, it is definitely not worth the battery drain. My battery drains a lot quicker if i am running at 748+ when the phone is on and being used, not idle and not in standby. I think I will be moving back down to 710 or 729.


*If you are not running the updated kernal and do not have JIT enabled like in the link above, your phone will have significantly less numbers being tested at same CPU speeds as mine listed.

great info, hoever you should add the disclaimer that most phones wont run up to 806 and some will crash at anything above 710. people tend to blame you for things like that.

you should do some side-by-side videos to show the speed difference.

also try running this test with other roms

i vote sticky btw

 

[thenestor]

Personally, I suspect that the JIT vs. non-JIT result is not much more than a pathology of that particular benchmark, and little more than that. I suppose I'd love to be proved wrong about that... because it would means that our little Eris keeps getting better and better with new software!

I was thinking about that too. The Linpack app might perform better with JIT, while other apps might not use JIT as effectively. I haven't heard any anecdotes like, "Such-and-such app runs well at 5 MFLOPS but didn't at 3 MFLOPS," so it doesn't seem particularly useful yet.

 

[reinert012]

great info, hoever you should add the disclaimer that most phones wont run up to 806 and some will crash at anything above 710. people tend to blame you for things like that.

you should do some side-by-side videos to show the speed difference.

also try running this test with other roms

i vote sticky btw

Hey thanks alot! I am definitely going to run some tests on some different ROMs hopefully soon if this gets stickied so everyone can see. Wont be a problem testing different ROMs, nandroid backups ftw.

 

[LexusBrian400]

Good info about JIT

Android Developers Blog: Dalvik JIT

 

[Caddyman]

I'm running ee @ 864 right now, no problems.

 

[reinert012]

I'm running ee @ 864 right now, no problems.

Some days i can run that high, but others i cant get past 810

 

[Caddyman]

after that i went to setcpu to run it back down a bit and the phone rebooted....lol

 

[dclay89]

806 is stable on mine but nothing after that..

 

[reinert012]

806 is stable on mine but nothing after that..

same here

 

[adammsu]

Agree with thenestor. I have seen a fair amount of "JIT make linpack run fast, but seems to have little actual impact." That is of course anecdotal....

 

[thenestor]

I feel like web browsing and keyboard response might be a little faster with JIT, but it might be a placebo. Other than those, I haven't perceived any difference in the apps that I personally use.

 

[ttran97]

What do you guys set your minimum at? For default and for the sleep profile? Thanks.

 

[ACD168]

I keep my normal profile at min-480/max-710
sleep profile at min-160/max-245
saves battery like crazy i go 4 days without charging. stock battery.

Running Ic3Romv2.9 maps.

 

[thenestor]

FWIW, I believe Zanfur mentioned that the stock CPU speeds are 245/245 asleep, 528/528 awake.

 

[reinert012]

FWIW, I believe Zanfur mentioned that the stock CPU speeds are 245/245 asleep, 528/528 awake.

Interesting didnt know, thanks!

 

[reinert012]

I keep my normal profile at min-480/max-710
sleep profile at min-160/max-245
saves battery like crazy i go 4 days without charging. stock battery.

Running Ic3Romv2.9 maps.

What exactly is maps? i hope its not something so obvious like google maps. I saw 2 downloads for 2.9, LWP and Maps, and just wasnt sure what maps was.

 

[ttran97]

What exactly is maps? i hope its not something so obvious like google maps. I saw 2 downloads for 2.9, LWP and Maps, and just wasnt sure what maps was.

Maps means there's Navigation on there and it works. There was a problem with Live Wallpapers not working with Maps, so there are two versions of Ic3...one with LWP, and one with Maps.

 

[reinert012]

Maps means there's Navigation on there and it works. There was a problem with Live Wallpapers not working with Maps, so there are two versions of Ic3...one with LWP, and one with Maps.

thanks for clearin that up! im loving ELB 2.4, just flashed it