ICS + Xoom Wifi Performance Changes

[Welch]

I decided to run some simple little benchmarks on my Xoom Wifi before and after the ICS 4.0.3 update. Here is what I found. Below is the raw data I got from running "Linpack for Android" and "Quadrant Standard". Both of these were run 3 times a piece (6 for Linpack to show Single/Multi thread). These are the raw numbers it put out and I'll leave them up to your own interpretation. Take them with a grain of salt considering that Linpack often returns a "Inconsistent" for "Precision", sort of annoying but that's why I listed the entire info as I saw it. You'll notice each first run of something generally takes longer, this should be due to that particular test being first ran and stored in the memory/cached. After that the numbers become a bit more consistent.

For those of you who just want averages, here they are. (I wouldn't put a lot of stock in them as too many "Inconsistent" tests were shown during these benchmarks).

***Before ICS***

Single Thread Average
MFLOPS: 32.034
Time: 2.62
Norm Res: 5.68

Multi-Thread Average
MFLOPS: 50.020
Time: 3.373
Norm Res: 3.24

Quadrant Before Average: 1659
-----------------------------------------

****After ICS****

Single Thread Average
MFLOPS: 29.177
Time: 2.883
Norm Res: 5.68

Multi-Thread
MFLOPS: 27.514
Time: 9.093
Norm Res: 3.24


Quadrant After Average: 1730.66
------------------------------------------------------------------





***BEFORE***
System:
SDK Version: 3.2.1
ID: HTK75D
OS name: Linux
OS Version: 2.6.36.3-g111d827
______________________________
LinPack

*****Single Thread*****

First Run
MFLOPS: 32.157
Time: 2.61
Norm Res: 5.68
Precision: Inaccurate Result

Second Run
MFLOPS:32.022
Time: 2.62
Norm Res: 5.68
Precision: 2.220446049250313E-16

Third Run
MFLOPS: 31.924
Time: 2.63
Norm Res: 5.68
Precision: 2.220446049250313E-16

*****Multi-Thread*****
First Run
MFLOPS: 49.622
Time: 3.4
Norm Res: 3.24
Precision: Inconsistent Result

Second Run
MFLOPS: 49.55
Time: 3.4
Norm Res: 3.24
Precision: 2.220446049250313E-16

Third Run
MFLOPS: 50.849
Time: 3.32
Norm Res: 3.24
Precision: 2.220446049250313E-16

*************************************
Quadrant

Before ICS
First Run 1466
Second Run - 1752
Third Run - 1759

_____________________________________________________________

***AFTER***
System:
SDK Version: 4.0.3
ID: IML77
OS name: Linux
OS Version: 2.6.39.4-gf03a679

*****Single-Thread*****
MFLOPS: 27.192
Time: 3.08
Norm Res: 5.68
Precision: 2.2204460492250313E

MFLOPS: 31.54
Time: 2.66
Norm Res: 5.68
Precision: Incosistent

MFLOPS: 28.799
Time: 2.91
Norm Res: 5.68
Precision: 2.220446049250313E-16

*****Multi-Thread*****
First Run
MFLOPS: 54.655
Time: 3.09
Norm Res: 3.24
Precision: Inconsistent

Second Run
MFLOPS: 13.975
Time: 12.07
Norm Res: 3.24
Precision: Inconsistent

Third Run
MFLOPS: 13.913
Time: 12.12
Norm Res: 3.24
Precision: 2.220446049250313E-16

********************************************
Quadrant

After ICS
First Run - 1539
Second Run - 1833
Third Run - 1820

[derFunkenstein]

This bears out with AnTuTu benchmarks on my Kindle, too. ICS is faster across the board, slightly faster in most subtests but the CPU subtests are much, much faster. I think Google's OS is finally maturing.

[Welch]

The first thing I noticed when watching Quadrant run (Before ICS) was that the shading effects really sucked. Once I ran it the next time (After ICS) I noticed that the shading effects had completely changed 100%! It was the equivalent of coming from DX7 to DX8.1 .

I'm assuming that the new ICS update includes some software set instructions that the previous Gingerbread OS didn't, thus unlocking the ability to much more accurately display shadows and the like.

I still have 1 more tablet here that has 2.6 on it. I will attempt to get better benchmarks (probably from another program) before and after an ICS update. I'll also see if I can get a video of what I'm talking about during the Quadrant graphic test before and after to show you the dramatic improvements in the way it looks.