Announcement

Collapse
No announcement yet.

Power Consumption & Thermal Testing With The Core i9 7900X On Linux

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

  • Power Consumption & Thermal Testing With The Core i9 7900X On Linux

    Phoronix: Power Consumption & Thermal Testing With The Core i9 7900X On Linux

    Following my initial Intel Core i9 7900X Linux benchmarks this week were questions raised about its power use and thermal efficiency. Here are some tests looking at those factors, including the performance-per-Watt.

    http://www.phoronix.com/vr.php?view=24873

  • #2
    i7-7900X‘s much faster than i7-5960X.

    Comment


    • #3
      Excellent article Michael and it's refreshing to see that you are one of only a small number of reviewers on the Internet who understands that efficiency is not merely looking at an abstract wattage number and saying that fewer watts under load is the same thing as "efficiency".

      Incidentally, for the Linux kernel compilation benchmark using your average power consumption numbers and time numbers I get:

      R7 1800X = 172.9 watts * 81.41 sec --> 14075 Joules (J)
      i7 7700K = 133.9 watts * 103.17 sec --> 13814 (J)
      i9 7900X = 234.3 watts * 50.59 sec --> 11854 (J) [Roughly 19% more energy efficient than the 1800X and 16.5% more efficient than the 7700K]

      So for all the talk about energy efficiency, even in a rough form with very early firmware the 7900X is already in the lead for energy efficiency in compiling (not to mention having excellent performance). That's even more impressive considering both the 1800X an the 7700K are on dual-memory channel platforms that consume less energy at idle.
      Last edited by chuckula; 06-29-2017, 11:30 AM.

      Comment


      • #4
        Originally posted by chuckula View Post
        Excellent article Michael and it's refreshing to see that you are one of only a small number of reviewers on the Internet who understands that efficiency is not merely looking at an abstract wattage number and saying that fewer watts under load is the same thing as "efficiency".

        Incidentally, for the Linux kernel compilation benchmark using your average power consumption numbers and time numbers I get:

        R7 1800X = 172.9 watts * 81.41 sec --> 14075 Joules (J)
        i7 7700K = 133.9 watts * 103.17 sec --> 13814 (J)
        i9 7900X = 234.3 watts * 50.59 sec --> 11854 (J) [Roughly 19% more energy efficient than the 1800X and 16.5% more efficient than the 7700K]

        So for all the talk about energy efficiency, even in a rough form with very early firmware the 7900X is already in the lead for energy efficiency in compiling (not to mention having excellent performance). That's even more impressive considering both the 1800X an the 7700K are on dual-memory channel platforms that consume less energy at idle.
        I suspect there exists thermal thottling for i9-7900X. It's more powerful than Ryzen 7 1800X, however it takes much more passes than Ryzen to finish the test suite.

        Comment


        • #5
          Originally posted by chuckula View Post
          Excellent article Michael and it's refreshing to see that you are one of only a small number of reviewers on the Internet who understands that efficiency is not merely looking at an abstract wattage number and saying that fewer watts under load is the same thing as "efficiency".
          Dude, there are several things wrong with this test. I'll tell you a few.

          1) The test meter has a reported accuracy of 1,5%. The results you listed for 1800x and 7700k are within that 1,5%. You can't really make conclusion like 'Roughly 19% more energy efficient than the 1800X and 16.5% more efficient than the 7700K' when the meter has an accuracy of 1,5%. You just can't. Fail.

          2) The PSU has different efficiency for different wattages. The difference can be few %. E.g. 90% efficiency for 130W load and 95% efficiency for 230W load. The test setup doesn't even list the PSU maker & model. Fail

          3) You get better results from the DC side of things. Unfortunately this can be a bit hard, but there are cheap current meters for even up to 20A on Ebay. The problem is, how to get this instrument between the secondary caps and the rails from MOSFETs.

          Comment


          • #6
            Originally posted by caligula View Post

            Dude, there are several things wrong with this test. I'll tell you a few.

            1) The test meter has a reported accuracy of 1,5%. The results you listed for 1800x and 7700k are within that 1,5%. You can't really make conclusion like 'Roughly 19% more energy efficient than the 1800X and 16.5% more efficient than the 7700K' when the meter has an accuracy of 1,5%. You just can't. Fail.

            2) The PSU has different efficiency for different wattages. The difference can be few %. E.g. 90% efficiency for 130W load and 95% efficiency for 230W load. The test setup doesn't even list the PSU maker & model. Fail

            3) You get better results from the DC side of things. Unfortunately this can be a bit hard, but there are cheap current meters for even up to 20A on Ebay. The problem is, how to get this instrument between the secondary caps and the rails from MOSFETs.
            EVGA SuperNOVA NEX750G. A good choice.
            Last edited by qsmcomp; 06-29-2017, 01:12 PM. Reason: My sorry.

            Comment


            • #7
              between this tests and the VRM cooking problems with most x299 motherboards + the BIOS bugs discovered + damn Those temps + the insane prices, I'll be waiting for ThreadRipper TYVM.

              Sure, skylake-X have some muscle but the platform is a nightmare and the price is way too high overall to make it attractive outside the Fanboys realm, it seems Intel forgot Ryzen exists and that it can do 80-90% of the job(core for core) at almost 1/3 of the price, I mean just with the CPU cost alone I can build a full Ryzen 7 1700 system with uber DDR4 3600 RAM, awesome PSU, cool AF case, overclock it add a damn cool aftermarket Noctua cooler and prolly add a cheap 580 or pre overclocked 570(when the mining phase passes ofc) and still I don't need to delide the CPU cuz Ryzen is properly soldered neither I have to get a freaking expensive uber custom watercooler with VRM support or risk otherwise cook the mobo LOL.

              I'm even confident that with a full 7900x system money(close to 3k with a proper 8+4 mobo, liquid cooling loop with VRM support, delid tooling + GPU) I can make 3 Ryzen 5(or 7+7+5) full builds, cluster them and trounce the 7900x in any productive task(note that I believe there is absolutely no reason to go x299 for gaming only systems at all), prolly using not much more power overall

              Comment


              • #8
                Originally posted by qsmcomp View Post

                Michael's using a cheap EVGA PSU with poor performance which is produced by HEC.
                It's a gold rated 750W power supply that seems to be relatively average in what you would expect for systems in this range. If Larabel had used some exotic high-end PSU I'm sure people would have complained about that too.

                https://www.newegg.com/Product/Produ...82E16817438027

                Comment


                • #9
                  Originally posted by caligula View Post

                  Dude, there are several things wrong with this test. I'll tell you a few.

                  1) The test meter has a reported accuracy of 1,5%. The results you listed for 1800x and 7700k are within that 1,5%. You can't really make conclusion like 'Roughly 19% more energy efficient than the 1800X and 16.5% more efficient than the 7700K' when the meter has an accuracy of 1,5%. You just can't. Fail.
                  So? An error margin of 1.5% that you cannot in any way show is biased in favor or against any CPU in there (because it's connected to the exact same PSU in both setups so the error pretty much cancels out) is irrelevant to these metrics and is a tiny portion of the differences in efficiency. You sound like some guy in traffic court whining that the radar gun had a calibration error of 2mph (and couldn't even prove the calibration error was biased against him) when you were cited as going 20 over the limit. You won't walk out happy.

                  Originally posted by caligula View Post
                  2) The PSU has different efficiency for different wattages. The difference can be few %. E.g. 90% efficiency for 130W load and 95% efficiency for 230W load. The test setup doesn't even list the PSU maker & model. Fail
                  So? It's a real-world test. Are you seriously upset about a 750W PSU gold rated being used in a real-world test with an 1800X system? Do you expect Larabel to swap out the PSU for different tasks based on how much power they consume?

                  Originally posted by caligula View Post

                  3) You get better results from the DC side of things. Unfortunately this can be a bit hard, but there are cheap current meters for even up to 20A on Ebay. The problem is, how to get this instrument between the secondary caps and the rails from MOSFETs.
                  That's about the only point that has some merit although you don't pay the power company based on the final DC efficiency.


                  Comment


                  • #10
                    Originally posted by chuckula View Post

                    It's a gold rated 750W power supply that seems to be relatively average in what you would expect for systems in this range. If Larabel had used some exotic high-end PSU I'm sure people would have complained about that too.

                    https://www.newegg.com/Product/Produ...82E16817438027
                    I'm sorry. It looks like a cheap EVGA 500 B however it's an EVGA NEX750G which is built by FSP. A good 750W PSU.
                    Last edited by qsmcomp; 06-29-2017, 01:13 PM.

                    Comment

                    Working...
                    X