UDOO CPU Performance Comparison

Discussion in 'General Discussion' started by DracoLlasa, Dec 28, 2013.

  1. DracoLlasa

    DracoLlasa UDOOer

    Joined:
    Oct 15, 2013
    Messages:
    419
    Likes Received:
    3
    This is an attempt to perform a standardized benchmarking of the UDOO in comparison to other development boards on the market (that i have access too). This was particularly challenging because most other development boards only have a single CPU Core, so finding a tool/process that would put them all on a level playing field was a bit challenging.

    So first i wanted to thank Lemele for turning me on to the Sysbench tool in this thread on UDOO Overclocking: viewtopic.php?f=2&t=596
    From this i was able to get a tool that was able to run with 1, 2, 3 or 4 threads allowing me to perform the same work leveraging all available CPU cores.

    The task for each was to calculate a specific amount of prime numbers. I did 3 separate batches, to calculate, 2K, 4K, and 6K prime numbers. On the UDOO boards i was able to do so using more than 1 thread. This helped show a real performance improvement when a task is allowed to leverage the additional cores. Too many of the other benchmarking tools were just doing the work using a single core and while even in those cases the UDOO performed well, it didn't give the UDOO a chance to show its full potential.

    At the end of this post you will find a link to the Google spreadsheet with all the results.

    Each test was performed on each board 5 times. then the average was calculated and put on the summary tab (tab 1)
    For example, the raspberry pi, calculated 4K prime numbers, 5 times, and the averages were documented int he results/statistics section
    For each test there were a series of stats that i highlighted on the results page.
    For those that like to see all of the data, you will find a tab for each board/setup. in the case of the RasPi, there are 3 tabs, one at default 700Mhz, and 2 others with it overclocked at 900Mhz and 1Ghz.
    In these tabs you will see the full results of each sysbench run, the commands are at the top of each column on the detailed tabs.

    I am aware this isn't a full 'real world' test, but until i can find better tools that can run on all the platforms this is the best i can produce now.
    Details of each board, and OS are noted on in the spreadsheet as well.

    As you will quickly see the UDOO dual and QUAD are faster by many orders of magnitude.

    Please feel free to post any questions, feedback or suggestions.
    Results Spreadsheet:
    http://goo.gl/rtwbLJ

    ToDo:
    - - Complete testing of BeagleBone Black

    Updates:
    - -
     
  2. EBrown

    EBrown New Member

    Joined:
    Dec 9, 2013
    Messages:
    48
    Likes Received:
    0
    We used to do a similar thing with programming contests. The challenge: Calculate the first 1,000,000 prime numbers.

    Another test I would do is floating-point arithmetic. Accuracy and speed are important on it. It will show the different speeds between the hard-float and soft-float implementations. (Which floating-point math is very processor consuming.)

    Another good test would be series/summation tests. Repetition among a process. (For example, Taylor Series would be a good test. Or even building a Mandelbrot would be a good test.)

    Then you would want to do some sort of graphics test. Do a 2D and a 3D test.

    Then the final test would be some sort of test for RAM speed. Load and unload a ----ton of data. Lot's of it. Load a 500MB file, AND a series of 10MB files.

    These are all going to give you good data to compare between the boards.

    Thanks,
    EBrown
     
  3. DracoLlasa

    DracoLlasa UDOOer

    Joined:
    Oct 15, 2013
    Messages:
    419
    Likes Received:
    3
    yes those are all good tests and things i have looked at. Doing them all on single core boards in feasible as there are documented ways of doing so. the problem is many of those things use binaries, or python scripts that only use a single core and as such dont allow the multi-core boards to show full potential. So anyways, as i noted your welcome to point me towards any muti-threaded/core capable testing or benchmarking tools. Until then this is the mot i can collect.

    The graphics performance is different and i still may be able to make some progress there, but i haven't put a lot of effort into that area just yet.
     
  4. EBrown

    EBrown New Member

    Joined:
    Dec 9, 2013
    Messages:
    48
    Likes Received:
    0
    Write your own tests. Just write them in C++ or Python and write them to be multi-core. I would, but my Udoo isn't working right now and I don't have enough time to dedicate to it.

    Thanks,
    EBrown
     
  5. DracoLlasa

    DracoLlasa UDOOer

    Joined:
    Oct 15, 2013
    Messages:
    419
    Likes Received:
    3
    i dont know how to write that kind of stuff.. im just starting to learn Python.. i can do "hello world" and thats about it :)

    Im def not an expert in this stuff, im just try hard to run meticulous tests with what resources i can find.
     
  6. marchdvd

    marchdvd New Member

    Joined:
    Dec 28, 2013
    Messages:
    4
    Likes Received:
    0
    I wrote a simple Python script to compare Beagle Bone Black, UDOO (quad) and Raspberry Pi and pcDuino and my iMac.
    see link http://marchdvd.com/?p=17
    iMac was 6 seconds UDOO was 30 seconds and others 53 seconds and over 2 minutes.
    I'm very impressed with the UDOO quad.
     
  7. DracoLlasa

    DracoLlasa UDOOer

    Joined:
    Oct 15, 2013
    Messages:
    419
    Likes Received:
    3
    all other python scripts i have used only leverage a single processor, does this one use more?
     

Share This Page