This page provides comparison among various TI processors running TI's Android-DevKit. Data published here is for the latest version of released Android-DevKit. Items yet to be measured are marked "TBD" and those that are not applicable to a particular platform are marked "N/A".
Platform Benchmarks
This is a collection of benchmarks comparing mainly CPU speed and other data transfer operations. In below tests, AM3715@1GHz, BeagleXM@1GHz, and AM3517@600MHz.
RowboPerf: ARM
See RowboPERF User Guide for more information about these metrics. The RowboPerf application calls native executables compiled with Android's NDK to perform these tests.
Numbers shown here measure the time from start of loading the kernel image to appearance of Android's Home Screen. These measurements are done using the default (pre-built) images provided with the DevKit. Please note that the very first time you boot your device, the boot time will be significantly higher (approx. 3-4 min).
Time (secs)
AM3715
BeagleXM
AM3517
SD First Boot Time1
64.8
70.5
SD Boot Time
29.4
32.9
1 First boot time refers to first time the system is booted with a freshly created SD card.
These tests were performed using the StorageIO application provided in RoboPerf. The application measures the write and read speed of different storage devices using several block sizes. For these tests the file size chosen was 100 MBytes. In order for the application to run the storage device chosen for the test must be mounted in the Android file system and a have read and write access. All the tests were perform on devices with a vfat file system.
MMC/SD Performance
These tests were performed on a SanDisk 2 GB, class 2, SD card. The third partition "mmcblk0p3" created on a bootable SD card as explained in http://processors.wiki.ti.com/index.php/TI-Android-GingerBread-2.3-DevKit-1.0_UserGuide#Procedure_to_populate_MMC.2FSD_Card was used to perform the IO operations. These results were obtained with a relatively new SD card with very few bad blocks, using SD cards of the same class with a considerable number of bad blocks will degrade the IO performance.
Throughput (MB/s)
AM3715
BeagleXM
AM3517
Write, bs=512
3.200
3.027
Write, bs=4096
3.686
3.240
Write, bs=16384
3.712
2.325
Write, bs=65536
3.878
2.475
Write, bs=524288
3.888
2.365
Write, bs=1048576
4.060
2.440
Read, bs=512
7.135
6.731
Read, bs=4096
7.829
9.235
Read, bs=16384
7.504
8.806
Read, bs=65536
7.709
9.030
Read, bs=524288
7.402
9.147
Read, bs=1048576
8.208
8.604
USB Performance
Throughput (MB/s)
AM3715
BeagleXM
AM3517
Write, bs=512
7.074
4.244
Write, bs=4096
9.271
4.638
Write, bs=16384
9.607
4.774
Write, bs=65536
9.311
4.928
Write, bs=524288
9.286
4.847
Write, bs=1048576
9.075
4.961
Read, bs=512
7.242
7.840
Read, bs=4096
10.018
9.268
Read, bs=16384
10.479
8.850
Read, bs=65536
10.424
9.118
Read, bs=524288
10.744
8.728
Read, bs=1048576
10.117
8.820
Connectivity Performance
These tests provide throughput comparison across commonly used transports/mediums.
Ethernet
Measures TCP Throughput between Server (running on Host PC) and Client (Android Platform) using different buffer sizes (bs in bytes) Both Server and client uses the Netperf utility. For more information about Netperf, please see Netperf Home page
TCP Throughput
AM3715
BeagleXM
AM3517
bs=16
18.66 Mb/s
Mb/s
33.48 Mb/s
bs=1024
18.65 Mb/s
Mb/s
33.87 Mb/s
bs=4096
48.89 Mb/s
Mb/s
49.94 Mb/s
bs=8192
54.85 Mb/s
Mb/s
11.87 Mb/s
Wireless LAN (WiFi)
Measures TCP Throughput between Server (running on Host PC) and Client (Android Platform) using different buffer sizes (bs in bytes) Both Server and client uses the Netperf utility. For more information about Netperf, please see Netperf Home page
Non-Secure
The CPU load shown is the highest number observed while running TOP for
the duration of the test.
AM3715
TCP Throughput
CPU Util (%)
bs=1024
5.67 Mb/s
20.56 % (tiwlan:16.89%,ti_sdio_drv:3.67%)
bs=4096
9.68 Mb/s
28.77% (tiwlan:24.33%,ti_sdio_drv:4.44%)
bs=8192
15.86 Mb/s
28.33% (tiwlan:19.33%,ti_sdio_drv:9%)
Secure
These tests were run with a buffer size of 4096 bytes (i.e. bs=4096).
The CPU load shown is the highest number observed while running TOP for
the duration of the test.
AM3715
TCP Throughput
CPU Util (%)
PSA-PSK Authentication
9.47 Mb/s
22.11 % (tiwlan:17.67%,ti_sdio_drv:4.44%)
WEP 64bit Encryption
9.93 Mb/s
23.56% (tiwlan:19%,ti_sdio_drv:4.56%)
WEP 128bit Encryption
9.88 Mb/s
23.66% (tiwlan:19.22%,ti_sdio_drv:4.44%)
ADB
Measure Android Debug Bridge (ADB) throughput between Host PC and Android device by pushing and pulling a 20MB file with random content 10 times. TX and RX numbers are from the Host PC point of view, so TX is measured by doing "adb push" on the Host PC.
Tests to measure video, audio, image and speech processing performance.
Video Display
To measure Video decode and display performance, the Big Buck Bunny clip was used. The clip has the following video and audio streams Stream #0.0(und): Video: mpeg4, yuv420p, 352x288 [PAR 48:11 DAR 16:3], 30 tbr, 30 tbn, 30 tbc Stream #0.1(und): Audio: aac, 48000 Hz, stereo, s16 DDR2 memory was set to 256M in the boot parameters (Default value)
For Android on DM37x EVM, which has multimedia DSP acceleration enabled, the sintel-1280-stereo.mp4 clip was used. The clip has the following video and audio streams. The ARM load (observed with 'top' command) is about 4~8% while playing the clip.
Video: AVC High@L3.1, CABAC, 5 ReFrames, 1,616 Kbps, 1280x544, 24.000 fps, yuv420p
Audio: AAC Ver4, LC, 192 Kbps, 2 channels, 48.0 KHz
Video decode
AM3715
BeagleXM
AM3517
DM3730
Min (fps)
AVG (fps)
Max (fps)
Power Management
To measure power consumption a Keithley 2000 multimeter with a multiple-channel scan card was used. vdd1 (core) and vdd2 (pheripherals) power was measured under different scenarios.
SUSPEND
Power consumption when the boards goes into SUSPEND mode.
Suspend Power sleep_while_idle=0,enable_off_mode=0
AM3715
vdd1 (mw)
9.23
vdd2 (mw)
54.87
vdd1+vdd2 (mw)
64.10
Suspend Power sleep_while_idle=1,enable_off_mode=0
AM3715
vdd1 (mw)
9.13
vdd2 (mw)
54.69
vdd1+vdd2 (mw)
63.82
Suspend Power sleep_while_idle=0,enable_off_mode=1
AM3715
vdd1 (mw)
7.03
vdd2 (mw)
52.82
vdd1+vdd2 (mw)
59.85
Suspend Power sleep_while_idle=1,enable_off_mode=1
