Asus TinkerBoard

[sfx2000]

This one is a bit under the radar...

http://www.slideshare.net/NiyaziSARAL/asus-tinker-board

Quad Rockchip Cortex-A17 in RPi 2/3 form factor...

Kinda puts them at odds with the Aeon UP board (Aeon is an Asus subsidiary) - the UP board being an X86 based device in similar form factor..

 

[sfx2000]

Tinkerboard is now available in the US over on the Amazon...

$60 is a bit higher than Pi3, but it's a stout board...

 

[System Error Message]

for 4 A17 cores, if its faster than the A15 than i definitely would give this a go. It even has gigabit ethernet too.

 

[sfx2000]

for 4 A17 cores, if its faster than the A15 than i definitely would give this a go. It even has gigabit ethernet too.

The challenge is the SW - but Asus seems to be making a good effort - not just Linux, but they've delivered an Android build-image...

The GPU is a bit problematic, as are all ARM chips running ARM Mali GPU's - but initial reviews are positive...

If the project doesn't need a GPU, then this is a very nice board to "Tinker" about with...

I'm just surprised that with the strong Asus contingent we have here on Smallnetbuilder, that this device doesn't get more attention.

It's a lot more stout and capable for linux tinkering than the typical Asus BHR's represented - might even make for a better VPN and other apps platform.

 

[System Error Message]

The challenge is the SW - but Asus seems to be making a good effort - not just Linux, but they've delivered an Android build-image...

The GPU is a bit problematic, as are all ARM chips running ARM Mali GPU's - but initial reviews are positive...

If the project doesn't need a GPU, then this is a very nice board to "Tinker" about with...

I'm just surprised that with the strong Asus contingent we have here on Smallnetbuilder, that this device doesn't get more attention.

It's a lot more stout and capable for linux tinkering than the typical Asus BHR's represented - might even make for a better VPN and other apps platform.

I checked wikipedia as it has a drystone mips rating. The A17 is faster than the A9 which is faster than A7 but is slower than the A15. However the A17 is slightly faster than the A53. As long as it maintains its current price i'll be picking one up.

 

[System Error Message]

upboard website not working, cant be a coincidence with this out if they are both asus

 

[sfx2000]

I checked wikipedia as it has a drystone mips rating. The A17 is faster than the A9 which is faster than A7 but is slower than the A15. However the A17 is slightly faster than the A53. As long as it maintains its current price i'll be picking one up.

A17 is a respin of A12 (which didn't get much traction as a 'big core' compared to A9/A15) - it's a good chip however, and one find this same chip in many Chromebooks and Kodi boxes...

(History here - A9 --> A12 --> A17 --> A73

A15 was a "big core" partner to A7, just as A53 is to A57)

Oddly enough - both are ARM's, but from different ARM design centers - A17 is Sophia, A15 is Cambridge...

Rockchip has their follow on design - the RK3399 - it's a Quad-A53/Dual-A72 big.LITTLE design with Mali T-860 GPU...

The 3399 is what Samsung is using in their recently release Chromebook Plus - and it's close performance-wise to the Broadwell-m3 in the Chromebook Pro.

 

[sfx2000]

This might better explain things... missing a couple of A's...

A73 fits into the A17 path... and A17 was a respin of A12 (replacing A9, which didn't happen)

A8 was directly replaced by A9 - A9 brought SMP into the house, A8 was the first out of order processor core.

The cortex family relationship is a bit odd, but it works... and then we have variants like Apple, Qualcomm, and Marvell... which may or may not be direct ARM equivalents.

 

[System Error Message]

A7 < A9 < A53 < A17 < A15 < A57
Thats the speed differences. A17 is smaller than A15, less performance but more efficient. Not something you'd want for your daily PC as the A15 is a better choice for that. But for EEE, prototyping and running headless they do fine.

 

[sfx2000]

A7 < A9 < A53 < A17 < A15 < A57
Thats the speed differences. A17 is smaller than A15, less performance but more efficient. Not something you'd want for your daily PC as the A15 is a better choice for that. But for EEE, prototyping and running headless they do fine.

Maybe, but things aren't quite that simple with ARM's and the SoC vendors...

Not just the cores, but the on-chip fabric and I/O - and consider that ARM has optional features...

A35 is the follow on to A7 as an example...

Then we have A72/A73 - competing designs that are the follow up from A15/A17 - which supplant A57...

Crazy stuff... and one looks at Intel - they were pretty clear, even AMD with their A series, which was a bit more homework to sort out what core was actually in play...

Risk was with AMD, bet wrong and choose the wrong chip, and one was seriously in trouble there... pick the right chip, and it was very, very good...

 

[sfx2000]

upboard website not working, cant be a coincidence with this out if they are both asus

Up now...

http://www.up-board.org/

 

[System Error Message]

ARM A15 is a lot faster than A17.

 

[sfx2000]

ARM A15 is a lot faster than A17.

a bit faster, at some things...

Application/Usage profiles... A17 is a nice chip - MIPS/Watt, and close to A15 there...

Anyways, these are both old-school, hard to believe, but they are...

 

[System Error Message]

a bit faster, at some things...

Application/Usage profiles... A17 is a nice chip - MIPS/Watt, and close to A15 there...

Anyways, these are both old-school, hard to believe, but they are...

Still both the A15 and A17 is faster than the A53 which is too common, only difference is the armv7 vs armv8.

 

[sfx2000]

Still both the A15 and A17 is faster than the A53 which is too common, only difference is the armv7 vs armv8.

A53 is comparable in performance to A9 - A17 is an evolution of A12, and developed alongside A15...

 

[sfx2000]

Quick notes on Tinker - this is with TinkerOS 2.03

cat /proc/cpuinfo
processor    : 0
model name    : ARMv7 Processor rev 1 (v7l)
BogoMIPS    : 10.08
Features    : half thumb fastmult vfp edsp thumbee neon vfpv3 tls vfpv4 idiva idivt vfpd32 lpae evtstrm 
CPU implementer    : 0x41
CPU architecture: 7
CPU variant    : 0x0
CPU part    : 0xc0d
CPU revision    : 1

openssl speed aes-128-cbc aes-256-cbc bf-cbc (common openssl benchmark thing)

TinkerBoard

The 'numbers' are in 1000s of bytes per second processed.
type             16 bytes     64 bytes    256 bytes   1024 bytes   8192 bytes  16384 bytes
blowfish cbc     54144.60k    59107.88k    60459.95k    60892.84k    61014.02k    60948.48k
aes-128 cbc      76739.59k    86328.03k    89633.93k    90330.07k    90476.69k    90608.98k
aes-256 cbc      59975.87k    63719.38k    65466.11k    66548.36k    66019.33k    66153.13k

Pi3

The 'numbers' are in 1000s of bytes per second processed.

type             16 bytes     64 bytes    256 bytes   1024 bytes   8192 bytes  16384 bytes
blowfish cbc     27761.14k    31865.22k    33075.63k    33421.31k    33521.66k    33527.13k
aes-128 cbc      40863.14k    46221.10k    48002.82k    48552.96k    48687.79k    48687.79k
aes-256 cbc      32267.21k    35372.18k    36294.74k    36563.97k    36640.09k    36651.01k

Sysbench (Tinker vs. Pi3)

sysbench --test=cpu --cpu-max-prime=20000 --num-threads=4 run
sysbench 0.4.12:  multi-threaded system evaluation benchmark

Running the test with following options:
Number of threads: 4

TinkerBoard

Doing CPU performance benchmark

Threads started!
Done.

Maximum prime number checked in CPU test: 20000


Test execution summary:
    total time:                          63.5912s
    total number of events:              10000
    total time taken by event execution: 254.3184
    per-request statistics:
         min:                                 25.08ms
         avg:                                 25.43ms
         max:                                 57.05ms
         approx.  95 percentile:              26.56ms

Threads fairness:
    events (avg/stddev):           2500.0000/8.86
    execution time (avg/stddev):   63.5796/0.01
   
Pi3

Maximum prime number checked in CPU test: 20000


Test execution summary:
    total time:                          92.4846s
    total number of events:              10000
    total time taken by event execution: 369.8503
    per-request statistics:
         min:                                 36.89ms
         avg:                                 36.99ms
         max:                                101.62ms
         approx.  95 percentile:              37.05ms

Threads fairness:
    events (avg/stddev):           2500.0000/3.94
    execution time (avg/stddev):   92.4626/0.01

 

[sfx2000]

And some visuals... as you can see - the floor plan is similar - lot more chips with the TinkerBoard - as it has a discrete GigE ethernet PHY which is directly support by the RK3288 - WiFi is via USB using the RT2800usb driver (Ralink/MediaTek) - Sound is also USB

Tinker is very sensitive to power issues - one really needs a robust 5VDC/2.4A power supply - and this is bare minimum - the office Raspberry Pi 2.5A PSU works well, and I've have good luck with CanaKit's 2.4A PSU - most of the issues with TinkerBoard have been power related - and this really comes down to the decision to put input power on a MicroUSB connector - CellPhone chargers here will not cut it as they can't deliver the power, and most cellphone charger cables, the AWG is too light to carry that much current. Pi3 has similar issues, but Tinker takes that one level beyond - must have a good power supply.

Software - Asus is pushing out some updates - community support is there - the Mali GPU is a bit of a challenge, but getting better - since RK3288 is also used with many Chromebooks/things - it does have some mainline Linux support compared to AllWinnner and other SoC's targeting the SBC market.

GPIO access - not as clean as RPi3 - but it is what it is - work is continuing there.

 

[sfx2000]

upboard website not working, cant be a coincidence with this out if they are both asus

The upboards are still avail - https://up-shop.org/4-up-boards

Intel has shifted gears as of late - killing off most of the IoT and maker stuff... Wonder where the uDoo stuff lands these days?

Intel's low-power stuff is mostly focused on the low-end NUC's and Chromebooks - and they do have Denverton, which is actually pretty interesting, but this is not a SBC kind of chip...

 

[sfx2000]

and just for fun...

https://github.com/kdlucas/byte-unixbench



Pi3 - Cortex-A53 @ 1.2GHz

------------------------------------------------------------------------
Benchmark Run: Sat Dec 30 2017 01:43:47 - 02:12:24
4 CPUs in system; running 4 parallel copies of tests

Dhrystone 2 using register variables       29690266.6 lps   (10.0 s, 7 samples)
Double-Precision Whetstone                     4687.2 MWIPS (11.8 s, 7 samples)
Execl Throughput                               3354.2 lps   (29.7 s, 2 samples)
File Copy 1024 bufsize 2000 maxblocks        235680.6 KBps  (30.0 s, 2 samples)
File Copy 256 bufsize 500 maxblocks           66923.2 KBps  (30.0 s, 2 samples)
File Copy 4096 bufsize 8000 maxblocks        614049.9 KBps  (30.0 s, 2 samples)
Pipe Throughput                             1057130.6 lps   (10.0 s, 7 samples)
Pipe-based Context Switching                 186745.9 lps   (10.0 s, 7 samples)
Process Creation                               8527.2 lps   (30.0 s, 2 samples)
Shell Scripts (1 concurrent)                   4590.3 lpm   (60.1 s, 2 samples)
Shell Scripts (8 concurrent)                    836.7 lpm   (60.1 s, 2 samples)
System Call Overhead                        2144664.0 lps   (10.0 s, 7 samples)

System Benchmarks Index Values               BASELINE       RESULT    INDEX
Dhrystone 2 using register variables         116700.0   29690266.6   2544.2
Double-Precision Whetstone                       55.0       4687.2    852.2
Execl Throughput                                 43.0       3354.2    780.1
File Copy 1024 bufsize 2000 maxblocks          3960.0     235680.6    595.2
File Copy 256 bufsize 500 maxblocks            1655.0      66923.2    404.4
File Copy 4096 bufsize 8000 maxblocks          5800.0     614049.9   1058.7
Pipe Throughput                               12440.0    1057130.6    849.8
Pipe-based Context Switching                   4000.0     186745.9    466.9
Process Creation                                126.0       8527.2    676.8
Shell Scripts (1 concurrent)                     42.4       4590.3   1082.6
Shell Scripts (8 concurrent)                      6.0        836.7   1394.5
System Call Overhead                          15000.0    2144664.0   1429.8
                                                                   ========
System Benchmarks Index Score                                         890.8

TinkerBoard - Cortex-A17 @ 1.8 GHz

------------------------------------------------------------------------
Benchmark Run: Fri Dec 29 2017 17:46:03 - 18:14:23
4 CPUs in system; running 4 parallel copies of tests

Dhrystone 2 using register variables       14109942.7 lps   (10.0 s, 7 samples)
Double-Precision Whetstone                     3740.9 MWIPS (11.2 s, 7 samples)
Execl Throughput                               1878.8 lps   (29.7 s, 2 samples)
File Copy 1024 bufsize 2000 maxblocks        173954.0 KBps  (30.0 s, 2 samples)
File Copy 256 bufsize 500 maxblocks           46495.3 KBps  (30.0 s, 2 samples)
File Copy 4096 bufsize 8000 maxblocks        448209.9 KBps  (30.0 s, 2 samples)
Pipe Throughput                              864653.9 lps   (10.0 s, 7 samples)
Pipe-based Context Switching                 177638.5 lps   (10.0 s, 7 samples)
Process Creation                               3964.6 lps   (30.0 s, 2 samples)
Shell Scripts (1 concurrent)                   3447.7 lpm   (60.1 s, 2 samples)
Shell Scripts (8 concurrent)                    428.8 lpm   (60.4 s, 2 samples)
System Call Overhead                        1892565.1 lps   (10.0 s, 7 samples)

System Benchmarks Index Values               BASELINE       RESULT    INDEX
Dhrystone 2 using register variables         116700.0   14109942.7   1209.1
Double-Precision Whetstone                       55.0       3740.9    680.2
Execl Throughput                                 43.0       1878.8    436.9
File Copy 1024 bufsize 2000 maxblocks          3960.0     173954.0    439.3
File Copy 256 bufsize 500 maxblocks            1655.0      46495.3    280.9
File Copy 4096 bufsize 8000 maxblocks          5800.0     448209.9    772.8
Pipe Throughput                               12440.0     864653.9    695.1
Pipe-based Context Switching                   4000.0     177638.5    444.1
Process Creation                                126.0       3964.6    314.7
Shell Scripts (1 concurrent)                     42.4       3447.7    813.1
Shell Scripts (8 concurrent)                      6.0        428.8    714.6
System Call Overhead                          15000.0    1892565.1   1261.7
                                                                   ========
System Benchmarks Index Score                                         605.7

Intel N3700 @ 1.6GHz

------------------------------------------------------------------------
Benchmark Run: Fri Dec 29 2017 18:15:12 - 18:43:07
4 CPUs in system; running 4 parallel copies of tests

Dhrystone 2 using register variables       50099027.1 lps   (10.0 s, 7 samples)
Double-Precision Whetstone                     8385.6 MWIPS (10.0 s, 7 samples)
Execl Throughput                               5704.7 lps   (30.0 s, 2 samples)
File Copy 1024 bufsize 2000 maxblocks        434275.6 KBps  (30.0 s, 2 samples)
File Copy 256 bufsize 500 maxblocks          115974.3 KBps  (30.0 s, 2 samples)
File Copy 4096 bufsize 8000 maxblocks       1034083.9 KBps  (30.0 s, 2 samples)
Pipe Throughput                             3169406.3 lps   (10.0 s, 7 samples)
Pipe-based Context Switching                 422774.1 lps   (10.0 s, 7 samples)
Process Creation                              11147.0 lps   (30.0 s, 2 samples)
Shell Scripts (1 concurrent)                   9981.8 lpm   (60.0 s, 2 samples)
Shell Scripts (8 concurrent)                   1378.3 lpm   (60.1 s, 2 samples)
System Call Overhead                        4149683.3 lps   (10.0 s, 7 samples)

System Benchmarks Index Values               BASELINE       RESULT    INDEX
Dhrystone 2 using register variables         116700.0   50099027.1   4293.0
Double-Precision Whetstone                       55.0       8385.6   1524.7
Execl Throughput                                 43.0       5704.7   1326.7
File Copy 1024 bufsize 2000 maxblocks          3960.0     434275.6   1096.7
File Copy 256 bufsize 500 maxblocks            1655.0     115974.3    700.8
File Copy 4096 bufsize 8000 maxblocks          5800.0    1034083.9   1782.9
Pipe Throughput                               12440.0    3169406.3   2547.8
Pipe-based Context Switching                   4000.0     422774.1   1056.9
Process Creation                                126.0      11147.0    884.7
Shell Scripts (1 concurrent)                     42.4       9981.8   2354.2
Shell Scripts (8 concurrent)                      6.0       1378.3   2297.1
System Call Overhead                          15000.0    4149683.3   2766.5
                                                                   ========
System Benchmarks Index Score                                        1655.6

 

[System Error Message]

The upboards are still avail - https://up-shop.org/4-up-boards

Intel has shifted gears as of late - killing off most of the IoT and maker stuff... Wonder where the uDoo stuff lands these days?

Intel's low-power stuff is mostly focused on the low-end NUC's and Chromebooks - and they do have Denverton, which is actually pretty interesting, but this is not a SBC kind of chip...

funny enough, i have the udoo x86 (with pentium), tinkerboard and pi 3.