October 18, 2019, 05:14:34 am

News:

Have you visited the Allwinner Chipset wiki? - http://linux-sunxi.org/


BENCHMARKS (C1 & C2 & CT & RPi & Mk802IV )

Started by arrhenius, February 02, 2014, 04:26:36 am

Previous topic - Next topic

arrhenius

February 02, 2014, 04:26:36 am Last Edit: February 02, 2014, 05:28:16 am by arrhenius
Hi:

I dont own all this boards but i have access to at least 1 of each, i have made some benchmarks on them and i would like to share the results, maybe some find then usefull .

If someone suggest me a new bechmark i will try to test it on the boards and add the results to the table.


The Contestants :

1. RaspBerry Pi : Broadcom BCM2835 - 1 Core 700MHZ - 256MB RAM

2. CubieBoard 1 : Allwinner A10 -1 Cores 1.0GHZ - 1GB Ram

3. CubieBoard 2 : Allwinner A20 -2 Cores 1.0GHZ - 1GB Ram

4. CubieTruck : Allwinner A20 -2 Cores 1.0GHZ - 2GB Ram

5. Mk802IV :  RockChip RK3188 - 4 Cores 1.6GHZ - 2GB Ram

6. GK802 :  freescale i.MX6q - 4 Cores  - 1GB Ram

The tests :

Test 1 : CPU

Command :sysbench --test=cpu --cpu-max-prime=20000 run

Test 2 : Mysql

Command : sysbench --test=oltp --oltp-table-size=1000000 --mysql-db=test --mysql-user=root --mysql-password=xxxx prepare

sysbench --test=oltp --oltp-table-size=1000000 --mysql-db=test --mysql-user=root --mysql-password=xxxxx --max-time=60 --oltp-read-only=on --max-requests=0 --num-threads=8 run


The results :











































BoardOS & kernelSysbench CPU TESTSysbench MYSQL test
RPiRaspbian not done yet 20.07 Transactions/s
CB1Cubieez & 3.4.61(pat)781 seconds60.83 Transactions/s
CB2Cubieez & original kernel720 seconds96.08 Transactions/s
CTCubieez & 3.4.75(pat)757 seconds111.75 Transactions/s
Mk802IVPicuntu & 3.0.36+378 seconds197.96 Transactions/s
GK802 Xubuntu 12.04729 seconds305.68 transaction/sec



CB1 CPU

CB1 MYSQL

gustagi

freescale i.MX6q GK802 xubuntu 12.04, 1G ram, quad  , 729sec, 305.68 transaction/sec
Intel B820  xubuntu 13.10 , 47sec, 608 trans/sec

arrhenius

Quote from: gustagi on February 02, 2014, 05:23:27 am
freescale i.MX6q GK802 xubuntu 12.04, 1G ram, quad  , 729sec, 305.68 transaction/sec
Intel B820  xubuntu 13.10 , 47sec, 608 trans/sec


thxs

the mysql benchmark on the freescale is quite impressive, whitch version is, have you tuned something?

mine:
mysql --version
mysql  Ver 14.14 Distrib 5.5.31, for debian-linux-gnu (armv7l) using readline 6.2

bh-whd

Quote from: arrhenius on February 02, 2014, 04:26:36 am
sysbench --test=cpu --cpu-max-prime=20000 run


Hi!

Can you please repeat the test on the CubieTruck with "--num-threads=2" ?

Thanks!
bh-whd





Anjin

February 16, 2015, 08:20:15 am #4 Last Edit: February 16, 2015, 08:46:21 am by Anjin
linaro@cubieboard4:~$ sysbench --test=cpu --cpu-max-prime=20000 run
sysbench 0.4.12:  multi-threaded system evaluation benchmark

Running the test with following options:
Number of threads: 1

Doing CPU performance benchmark

Threads started!

Done.

Maximum prime number checked in CPU test: 20000


Test execution summary:
    total time:                          274.6984s
    total number of events:              10000
    total time taken by event execution: 274.6895
    per-request statistics:
         min:                                 27.44ms
         avg:                                 27.47ms
         max:                                 33.61ms
         approx.  95 percentile:              27.48ms

Threads fairness:
    events (avg/stddev):           10000.0000/0.00
    execution time (avg/stddev):   274.6895/0.00



And just for giggles....

Number of threads: 4
Test execution summary:
    total time:                          70.3334s
    total number of events:              10000
    total time taken by event execution: 281.2603
    per-request statistics:
         min:                                 27.44ms
         avg:                                 28.13ms
         max:                                 74.83ms
         approx.  95 percentile:              31.17ms

Threads fairness:
    events (avg/stddev):           2500.0000/1.22
    execution time (avg/stddev):   70.3151/0.01


Number of threads: 8
Test execution summary:
    total time:                          52.1474s
    total number of events:              10000
    total time taken by event execution: 416.8649
    per-request statistics:
         min:                                 27.44ms
         avg:                                 41.69ms
         max:                                167.57ms
         approx.  95 percentile:              62.88ms

Threads fairness:
    events (avg/stddev):           1250.0000/423.77
    execution time (avg/stddev):   52.1081/0.02

Tokka

Just for curiosity i ran the cpu test on my cb1, with 2 mediaserver running, remote desktop, and some other apps (approx near 350-400mb of ram used).
The result is this:
QuoteMaximum prime number checked in CPU test: 20000


Test execution summary:
    total time:                          648.0215s
    total number of events:              10000
    total time taken by event execution: 647.9880
    per-request statistics:
         min:                                 60.17ms
         avg:                                 64.80ms
         max:                                356.29ms
         approx.  95 percentile:              74.62ms

Threads fairness:
    events (avg/stddev):           10000.0000/0.00
    execution time (avg/stddev):   647.9880/0.00

#####  this one is with mediaservers stopped #####

Maximum prime number checked in CPU test: 20000


Test execution summary:
    total time:                          604.9585s
    total number of events:              10000
    total time taken by event execution: 604.9318
    per-request statistics:
         min:                                 59.92ms
         avg:                                 60.49ms
         max:                                123.95ms
         approx.  95 percentile:              62.26ms

Threads fairness:
    events (avg/stddev):           10000.0000/0.00
    execution time (avg/stddev):   604.9318/0.00



What do you think about the result?
CB1 (A10) - Qbee-X_TMC

Anjin

And while I was at it, I pulled the same for a Tegra K1....  to really clock a jetson, one should be using the GPU... :)

Test execution summary:
    total time:                          47.9930s
    total number of events:              10000
    total time taken by event execution: 191.8939
    per-request statistics:
         min:                                 18.98ms
         avg:                                 19.19ms
         max:                                144.44ms
         approx.  95 percentile:              19.17ms

Threads fairness:
    events (avg/stddev):           2500.0000/9.30
    execution time (avg/stddev):   47.9735/0.01

rgmhtt

This was just posted on the google cubieboard group:

On 02/17/2015 06:46 AM, Wilfried Gösgens wrote:
> I did a performance comparison of the multiple available locking techniques.
> The cubie truck did hold up against the rest pretty well:
>
> https://www.arangodb.com/2015/02/16/comparing-atomic-mutex-rwlocks
>
> Cheers,
> Willi
> --
> You received this message because you are subscribed to the Google Groups "Cubieboard" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to cubieboard+unsubscribe@googlegroups.com.
> Visit this group at http://groups.google.com/group/cubieboard.
> For more options, visit https://groups.google.com/d/optout.

Anjin

Quote from: Tokka on February 16, 2015, 03:21:36 pm
Test execution summary:
    total time:                          648.0215s
    total time taken by event execution: 647.9880
         min:                                 60.17ms
         avg:                                 64.80ms
         max:                                356.29ms

#####  this one is with mediaservers stopped #####
Test execution summary:
    total time:                          604.9585s
    total time taken by event execution: 604.9318
    per-request statistics:
         min:                                 59.92ms
         avg:                                 60.49ms
         max:                                123.95ms



What do you think about the result?
[/quote]

Tokka,

Benchmarking should be run in an idle environment.  I reduced the data quoted for your system. 

The total time is higher while running other apps, as they all wake up on occasion, and as such steal just a wee bit of time each.  Think of it as most will just wake up, process an interrupt or check some storage... go right back to sleep.  Since each app in the process tree might have differing tasks, demands are different.  So in both cases, the best test case took about 60ms to run.  That means when the wee little cpu could focus on just solving that one task, it got it done in around 60ms.

But, when the OS chopped up the task so it could share cpu with the extras you had running, it took a whopping 356ms!  So under some minimal load, im sure the apps were idle when you tested... you can expect the same unit of work, to consume five times as much time to complete.  In the worst case, after 10,000 tries... it COULD take five times as long.  On average, it took just 65ms to complete, vs 60ms without the apps, costing a total of 42 seconds over the course of 10,000 tries.

Thus it means the end of the universe is near, for a wee little A10 CPU on a Cubieboard off in the far corners of the internet is near to solving the question for the answer!  I do hope there will be fish.....