Page 5 of 5

Re: Android benchmark. Looking for testers

Posted: Mon Sep 24, 2012 9:15 pm
by sluggo
Running cm9 on a Touchpad, if that's of any interest. PM sent.

Edit ... ran it already. On a 1.2ghz touchpad, 992.0 on two threads.

Re: Android benchmark. Looking for testers

Posted: Mon Sep 24, 2012 9:57 pm
by codedivine
sluggo wrote:
Running cm9 on a Touchpad, if that's of any interest. PM sent.

Edit ... ran it already. On a 1.2ghz touchpad, 992.0 on two threads.


Thanks :)
Another result: A Galaxy S2 (exynos 4 dual-core, 1.2ghz) user reported 998 MFlops using 2 threads. Not sure of his firmware version though.

Re: Android benchmark. Looking for testers

Posted: Tue Sep 25, 2012 9:35 am
by codedivine
I think after looking at the data so far, we can conclude the following: Cortex A9 implementations such as Exynos 4 and OMAP4 are achieving about 0.4 flops/cycle on the multithreaded mode. Snapdragon S3 is also doing 0.4 flops/cycle.

However, Tegra 3 to be the exception to rule and is only achieving about 0.32 flops/cycle average. Tegra 2 is also stuck at about 0.36 flops/cycle. Wonder if it has something to do with Nvidia's memory controller.

About the single threaded results, actually I had forgotten that the single-threaded and multi-threaded versions are working on different problem sizes. The single threaded one is working on smaller matrices. This was to ensure that the single thread case does not take very long to run, but now starting to think that was a poor design decision on my part. So results from single threaded and multithreaded are not directly comparable. I think I will provide settings to choose the matrix sizes yourself sometime.

Re: Android benchmark. Looking for testers

Posted: Tue Sep 25, 2012 11:05 am
by ChronoReverse
Huh, I wonder if the VFP for the A9 has its own strengths compared to the S3. I wonder which would be more applicable in real life.

Re: Android benchmark. Looking for testers

Posted: Tue Sep 25, 2012 11:32 am
by willyjwebb
current version of rgbench from play store on galaxy s2 i9100m 915 mflops 2 thread. 4.1.1 rom
367 single thread

Re: Android benchmark. Looking for testers

Posted: Tue Sep 25, 2012 12:09 pm
by killadark
Is it just me or its the performance gone down
1548mflops for galaxy s3 international
4 threads

Re: Android benchmark. Looking for testers

Posted: Tue Sep 25, 2012 12:18 pm
by codedivine
killadark wrote:
Is it just me or its the performance gone down
1548mflops for galaxy s3 international
4 threads


Thanks. Well, the performance has not gone down, it is actually corrected :oops:
The v1.2 you tested earlier had a bug. The Play store version is the corrected version, and is the actual correct number.

Re: Android benchmark. Looking for testers

Posted: Tue Sep 25, 2012 12:22 pm
by codedivine
codedivine wrote:
killadark wrote:
Is it just me or its the performance gone down
1548mflops for galaxy s3 international
4 threads


Thanks. Well, the performance has not gone down, it is actually corrected :oops:
The v1.2 you tested earlier had a bug. The Play store version is the corrected version, and is the actual correct number.


Actually, still, the difference in perf is quite large. DId not expect that big a difference. Something not quite right, would you mind running it again?

Re: Android benchmark. Looking for testers

Posted: Tue Sep 25, 2012 12:33 pm
by killadark
Actually I think this might have something to do with my under volted KernAl will flash jelly bean tomorrow and rerun this
I just re ran it on ova with UV
1thread 458
4 threads 1961,1774 mflops
Kinda weird :-?

Re: Android benchmark. Looking for testers

Posted: Tue Sep 25, 2012 12:58 pm
by codedivine
Thanks. Different firmware versions can indeed have various effects on thread scheduling and frequency scaling, so might be the issue.
Here is my blogpost with prelim analysis of data, including assembler generated by GCC for the innermost loop. More technical readers will be interested: http://codedivine.org/2012/09/25/prelim ... rgbenchmm/

Re: Android benchmark. Looking for testers

Posted: Tue Sep 25, 2012 1:05 pm
by ChronoReverse
https://play.google.com/store/apps/deta ... lity&hl=en

StabilityTest has a natively compiled version of Linpak (the normal Linpak for Android is dalvik-based and thus even more terrible). It would be interesting to compare the results a bit.

Galaxy Note Snapdragon S3 @ 1.5GHz on Android 4.1 (JB, CM10)
1 thread: 100.2
2 threads: 210.4

Re: Android benchmark. Looking for testers

Posted: Tue Sep 25, 2012 1:18 pm
by codedivine
ChronoReverse wrote:
https://play.google.com/store/apps/details?id=com.into.stability&hl=en

StabilityTest has a natively compiled version of Linpak (the normal Linpak for Android is dalvik-based and thus even more terrible). It would be interesting to compare the results a bit.

Galaxy Note Snapdragon S3 @ 1.5GHz on Android 4.1 (JB, CM10)
1 thread: 100.2
2 threads: 210.4


Interesting. Was not aware of that test. The MFlop numbers are still not reflecting what should be possible with the processors. LINPACK is a somewhat misleading benchmark name. Linpack tests are actually not a single benchmark. Rather, it is more of a "Calculate this using whatever algorithm you feel appropriate, as long as results are accurate and stable". Most Linpacks on servers are rewritten to heavily use the BLAS, and heavily depend on good matrix multiplication performance to the point where Linpack performance is close to the flops in a matrix multiplication benchmark (such as mine).

However, reference Linpack implementations in C/C++ also exist. These are more for reference/learning and do not use the BLAS and not used to test performance on servers. Given the relatively low FLOP rating you got, I think perhaps what they have compiled is this reference version.

edit: However, it should still be a good benchmark and hopefully not as terrible as the Java version.

Re: Android benchmark. Looking for testers

Posted: Fri Sep 28, 2012 3:35 am
by killadark
killadark wrote:
Actually I think this might have something to do with my under volted KernAl will flash jelly bean tomorrow and rerun this
I just re ran it on ova with UV
1thread 458
4 threads 1961,1774 mflops
Kinda weird :-?


I'm getting the same result after flashing jelly bean so I think it's something to do with the benchmark

Re: Android benchmark. Looking for testers

Posted: Fri Sep 28, 2012 11:28 am
by codedivine
killadark wrote:
killadark wrote:
Actually I think this might have something to do with my under volted KernAl will flash jelly bean tomorrow and rerun this
I just re ran it on ova with UV
1thread 458
4 threads 1961,1774 mflops
Kinda weird :-?


I'm getting the same result after flashing jelly bean so I think it's something to do with the benchmark


Thanks. Hmm.

Re: Android benchmark. Looking for testers

Posted: Fri Sep 28, 2012 10:47 pm
by codedivine
Got result from a user with a Snapdragon S4 dual-core: About 1480 MFlops, matching that of Tegra 3 and about 25% better than the S3 at same clock speeds. Very impressive.

Re: Android benchmark. Looking for testers

Posted: Mon Oct 08, 2012 10:00 am
by drkicker
HTC One S - stock kernel and speed
1 thread - 749 MFlops
2 threads - 1486 MFlops
4 threads - 1508 MFlops