Announcement

Collapse
No announcement yet.

A Quick Test Of NVIDIA's "Carmel" CPU Performance

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #11
    Thanks And that's without updating the compiler, I'd expect gcc 8 to bring some significant improvements.

    I ran everything through OpenBenchmarking. That was quite easy. Great work, Michael

    OpenBenchmarking.org, Phoronix Test Suite, Linux benchmarking, automated benchmarking, benchmarking results, benchmarking repository, open source benchmarking, benchmarking test profiles


    My Jetson TX2 is faster than Jetson Xavier for all of the tests which makes me wonder if, like for TX2, the SoC is somehow not in the fastest mode by default.

    Comment


    • #12
      Originally posted by ldesnogu View Post
      My Jetson TX2 is faster than Jetson Xavier for all of the tests which makes me wonder if, like for TX2, the SoC is somehow not in the fastest mode by default.
      Yeah. If Nvidia is going to the trouble and expense of sending you another Jetson board, Michael, maybe they will tell you how to configure it for max performance (if you ask).

      BTW, I remember seeing different performance numbers for different TDP envelopes, on the TX2. I wonder if this has anything to do with it.

      Comment


      • #13
        Originally posted by phoronix View Post
        ...
        Beyond ldesnogu 's point, I wish you could add the nearest laptop & desktop CPUs to these graphs. They would provide a useful basis for comparison.

        Thanks!

        Comment


        • #14
          Originally posted by coder View Post
          BTW, I remember seeing different performance numbers for different TDP envelopes, on the TX2. I wonder if this has anything to do with it.
          You can find some information here: https://devblogs.nvidia.com/jetson-t...lligence-edge/

          Comment


          • #15
            Originally posted by coder View Post
            Beyond ldesnogu 's point, I wish you could add the nearest laptop & desktop CPUs to these graphs. They would provide a useful basis for comparison.
            It's pretty easy to do: install openbenchmarking and run phoronix-test-suite benchmark 1809258-RA-1809248RA57

            Against a Lenovo T480s with an i7-8650U (4 cores with HT, 1.9-4.2 GHz) https://openbenchmarking.org/result/...RA-1809258RA90

            Comment


            • #16
              Any estimation on Carmel IPC compared to Denver?

              Comment


              • #17
                More results for TX2: https://openbenchmarking.org/result/...RA18&obr_sor=y

                Comment


                • #18
                  Originally posted by ldesnogu View Post
                  Thanks, but why include an in-order single-core Atom CPU from 2010?

                  It would be much more informative and useful to compare with a modern, low-power core, like Goldmont. It's present in the Apollo Lake SoC, which can be found in the CompuLab Fitlet 2 that Michael tested, here:

                  https://www.phoronix.com/scan.php?pa...pulab-fitlet-2

                  Sadly, I don't see much overlap between those results. Of particular interest are compute-bound tests, which then limits the useful overlap to x264 and FLAC. CRay, as well, but I'm not sure the Fitlet 2 CRay test is comparable to the test case that you used.

                  Anyway, a Gemini Lake SoC, with its Goldmont+ cores, would be even better.
                  Last edited by coder; 27 September 2018, 04:45 AM.

                  Comment


                  • #19
                    Originally posted by coder View Post
                    Thanks, but why include an in-order single-core Atom CPU from 2010?
                    I picked the most recent tag with ARM results which happens to have been made by someone who benchmarked an Atom CPU. I am more interested in comparing ARM vs ARM. As a reminder, I did most of these tests to show the Jetson TX2 Micheal has used is not properly configured.

                    It would be much more informative and useful to compare with a modern, low-power core, like Goldmont. It's present in the Apollo Lake SoC, which can be found in the CompuLab Fitlet 2 that Michael tested, here:

                    https://www.phoronix.com/scan.php?pa...pulab-fitlet-2

                    Sadly, I don't see much overlap between those results. Of particular interest are compute-bound tests, which then limits the useful overlap to x264 and FLAC. CRay, as well, but I'm not sure the Fitlet 2 CRay test is comparable to the test case that you used.

                    Anyway, a Gemini Lake SoC, with its Goldmont+ cores, would be even better.
                    If I had free time and access to more HW, I happily would do all of that. Also my TX2 hasn't a HD/SSD so some of the tests will be meaningless.

                    Comment


                    • #20
                      Originally posted by ldesnogu View Post
                      I picked the most recent tag with ARM results which happens to have been made by someone who benchmarked an Atom CPU. I am more interested in comparing ARM vs ARM. As a reminder, I did most of these tests to show the Jetson TX2 Micheal has used is not properly configured.


                      If I had free time and access to more HW, I happily would do all of that. Also my TX2 hasn't a HD/SSD so some of the tests will be meaningless.
                      Thanks for doing what you did.

                      I was sort of just wondering if it's possible to add his Fitlet 2 results to your comparison. I've never submitted results, so I don't know exactly how it works.

                      Comment

                      Working...
                      X