Announcement

Collapse
No announcement yet.

A Look At The Per-Clock Performance / Peak Frequencies With The Intel Core i7-1065G7 Ice Lake

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

  • A Look At The Per-Clock Performance / Peak Frequencies With The Intel Core i7-1065G7 Ice Lake

    Phoronix: A Look At The Per-Clock Performance / Peak Frequencies With The Intel Core i7-1065G7 Ice Lake

    Following this week's Intel Core i7-1065G7 Ice Lake Linux benchmarks there was some questions and speculations about the per-clock performance of this long-awaited Intel microarchitecture update. Here is some additional data shedding light on the clock frequencies under load and ultimately how the per-clock performance compares to the common Intel previous-generation mobile CPUs.

    http://www.phoronix.com/vr.php?view=28419

  • #2
    Looks like Intel was "slightly" exaggerating the IPC gains of Icelake, to say the least.

    Comment


    • #3
      Typo:

      Originally posted by phoronix View Post
      First up was the single-threaded PHPBench test case where the Core i7-1065G7 came out to be 1.17x the spedd of the Core i7-8550U.

      Comment


      • #4
        Originally posted by brent View Post
        Looks like Intel was "slightly" exaggerating the IPC gains of Icelake, to say the least.
        Aren't you reading the graphs the wrong way around?
        I mean if you are reading them as Ice Lake hitting higher avg then IPC could be the same and the unit have higher performance.

        Ice Lake settles at a lower avg. lower min and lower max. Yet it still manages to beat it's older sibling handily.
        I can't interpret it any other way than higher IPC. Significantly so.

        Comment


        • #5
          Michael
          But in this set of tests if looking at the geometric mean of the raw performance, the Ice Lake CPU was running at 1.1x the speed of the Kabylake-R model. So ultimately the i7-1065G7 had about 1.08x the performance-per-MHz of the Kabylake-R model.
          that cant possibly be right
          1,1X perf
          running at lower clocks
          1,08X perf per MHz

          did you mean 1,18?

          EDIT: i did the math and its indeed 18% higher, matching intels claim
          Last edited by davidbepo; 10-26-2019, 12:25 PM.

          Comment


          • #6
            Originally posted by brent View Post
            Looks like Intel was "slightly" exaggerating the IPC gains of Icelake, to say the least.
            no, see my comment

            Comment


            • #7
              Originally posted by davidbepo View Post
              Michael


              that cant possibly be right
              1,1X perf
              running at lower clocks
              1,08X perf per MHz

              did you mean 1,18?

              EDIT: i did the math and its indeed 18% higher, matching intels claim
              Yeah fixed, thanks, meant 18 not 08 typo.
              Michael Larabel
              http://www.michaellarabel.com/

              Comment


              • #8
                Thank you, Michael

                It would be great if you also run something more computationally intensive on these two laptops, e.g. x264/x265 veryslow preset encoding or rendering in Blender - all of which should take at least a few minutes to complete. This way we'd get a little bit more consistent frequencies, temperatures, throttling and performance - not some spikes which don't convey a lot of information.

                Comment


                • #9
                  Originally posted by birdie View Post
                  Thank you, Michael

                  It would be great if you also run something more computationally intensive on these two laptops, e.g. x264/x265 veryslow preset encoding or rendering in Blender - all of which should take at least a few minutes to complete. This way we'd get a little bit more consistent frequencies, temperatures, throttling and performance - not some spikes which don't convey a lot of information.
                  There are some long running tests if you click the OpenBenchmarking.org link in the article. Also, each test ran a minimum of 3 minutes total, granted many tests longer than that, so at a minimum are 180 samples per test.
                  Michael Larabel
                  http://www.michaellarabel.com/

                  Comment


                  • #10
                    Hi Michael,

                    Great article, as usual, but I can't find memory speeds anywhere. What memory speeds are the two systems running at?

                    Comment

                    Working...
                    X