2 Replies Latest reply on Feb 17, 2015 8:03 PM by HughTay

    Benchmarking - Core i5 (4th Gen) vs. Core M

    HughTay Green Belt

      Good day everyone,

       

      We performed some benchmarking tests on 2 tablet engineering samples that we have, 1 equipped with the Core i5 4202Y and the other Core M 5Y10c, and we thought we'd share our data since it might be informative to anyone else also interested in comparing the 2 architectures.

       

      Both tablet samples have the same specs otherwise:

      - 4 GB RAM

      - 64 GB SSD

      - 1920x1080 LCD touchscreen

      - 6000 mAH battery

       

      In both cases the OS used is Fedora 21 with Gnome 3.14 user environment and Linux kernel v3.17.8.

       

      ~~~

       

      I've attached the benchmarking results for quantitative comparison, so I will just briefly summarize:

       

      General performance -

      Both the Core i5 and Core M models performed very well in start up & shutdown times, far better than most Android devices and comparable with high-end Windows 8.1 laptops/desktops. Battery life can't beat some of the energy-saving tablets out there like the iPad Air, but still beats out most laptops.

       

      Web browsing -

      Again both Core i5 and Core M models rendered web pages & dynamic HTML5 content very well, with little difference from each other, and significantly ahead of tablets & smartphones even with quad-core ARM chips.

       

      Development/Scientific/Engineering -

      This is where the Core i5 model has a visible lead over the Core M model. Multi-threading capabilities (e.g. compiling Linux on 8 threads) on the Core i5 are around 30% better than on the Core M, with similar results for scientific/engineering applications as shown by the CERN ROOT benchmark (which tests things like random number generation, curve-fitting, pattern-generation & Monte Carlo simulation).

       

      3D Graphics -

      To be honest this is where both models delivered only mediocre performance... integrated graphics has never been Intel's core strength I suppose. Still the Core i5 model again showed visible advantages over the Core M model in rendering 3D objects in OpenGL.

       

       

      Other issues that emerged during our testing:

      - Core i5's 7 hour battery life compared to Core M's 6 hours
      - Core M's touchscreen doesn't work after waking from sleep

       

      I believe these 2 issues are due to the Linux kernel not updated with the latest Intel ACPI drivers for Core M yet, leaving the OS to use the older ACPI drivers that aren't fully compatible with Core M. Given Intel's past and present contributions to Linux, I also believe it wouldn't take long for Intel to fix these issues, am I correct?


      Overall evaluation:

      The Core M tablet wins out in price and weight, while the Core i5 tablet wins out in performance and track record.

      (Multi-tasking & virtualization performance weren't benchmarked comprehensively enough at the time of writing to be included here.)

       

      ~~~

       

      If anyone has any questions about our methodology or updates about Core M support in Linux or know of any methods for us to improve the results, please feel free to share with us.

       

      We hope you found the info here helpful! =)

       

      Hugh Tay

        • Re: Benchmarking - Core i5 (4th Gen) vs. Core M
          gabriel.thomas Brown Belt

          Hello Hugh Tay.

           

          Thank you very much for sharing this information with the Community. We appreciate these valuable inputs to the forum, it is very helpful.

           

          Best Regards.

           

          Gabriel Thomas.

          • Re: Benchmarking - Core i5 (4th Gen) vs. Core M
            HughTay Green Belt

            Update for Virtualization Benchmarking

             

             

            We ran 2 simple tests on both the Core i5 and Core M tablet platforms to get a quick glimpse into how well virtual machines can perform on both platforms. The 1st test was a simple Python script in virtualized Debian (Live image on Virt-Manager and installed image on VirtualBox, 2 vCPU's & 1 GB RAM in both cases) which generated & sorted 100,000 random numbers. The 2nd test was Geekbench 3 on virtualized Android (v5.0.1 32-bit) with a 4" 480x800 display, 512 MB RAM & 32 MB heap.

             

            ---

             

            Core i5 results:

             

            VirtualBox Debian (Installed)-Real: 3.276s,User: 0.092s,System: 1.492s
            Virt-Manager Debian (Live CD)-Real: 3.182s,User: 0.332s,System: 0.288s

             

            Geekbench single-core score on Android Lolipop = 1651

             

            ---

             

            Core M results:

             

            Virt-Manager Debian (Live CD)-Real: 3.162s,User: 0.348s,System: 0.332s

             

            Geekbench single-core score on Android Lolipop = 1458


            ---

             

            Comments:

             

            We were unfortunately unable to actually get the VirtualBox VM running on the Core M platform though; the VM never made it past the loading dialogue box... If anyone can suggest a solution we'd be delighted to try out.

             

            For the virtualized Android benchmarking test, the Geekbench multi-core score was discarded since the default Android SDK emulator for x86 doesn't support >1 vCPU out of the box on Linux. If you need to speed up the emulator, this guide might be helpful.