Announcement

Collapse
No announcement yet.

Initial AMD Ryzen 7 4700U Linux Performance Is Very Good

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

  • #11
    Michael Is it possible to change TDP up in this laptop? So you can test it at 25W

    Also, there are rumors related to Ryzen 9 4900U, so there are plenty perf to look around from AMD part

    Comment


    • #12
      81YM0002US is sold out, and I thought they were going to offer a fedora version...

      Comment


      • #13
        I just ordered this exact laptop this weekend. It's scheduled to arrive on Thrusday! I can't wait to try out Fedora on it.

        Comment


        • #14
          I wish they would make better keyboards for laptops. At least there seem the be less glare and more IPS >1366x768 screens these days. But since a notebook is a mobile typewriter for me mainly (and maybe do some darktable and stuff) I really need a good keyboard (layout and key haptics).
          Stop TCPA, stupid software patents and corrupt politicians!

          Comment


          • #15
            4900hs laptop https://altex.ro/laptop-gaming-asus-...PGA401IVHA033/

            Comment


            • #16
              I have the IdeaPad 15" Ryzen 5 3500U and it has some issues with some of the more newer kernels with Arch & Clear Linux.

              For reasons I haven't been to sort, the later kernels cant initiate the IOMMU required for AMD-Vi support. No AMD-Vi, no KVM.

              All the later Ubuntu builds, no problem. So hopefully its a one time issue with a model and not prevalent in the AMD Mobile Ryzen world.

              Comment


              • #17
                Originally posted by edwaleni View Post
                For reasons I haven't been to sort, the later kernels cant initiate the IOMMU required for AMD-Vi support.
                If you can, try building Linus' master branch; I swear I'd just seen a number of AMD IOMMU changes go in over the last few days.

                Comment


                • #18
                  Numpy uses a lot of C extensions. Was it compiled from source or does the Python package manager just pull in the generic Linux binary blob they have? It could be that the default Numpy binary is optimized for Intel. Some python distributions have offered numpy with a BLAS library compiled with Intel's compiler, known to use no optimizations if it detects an AMD CPU.

                  Comment


                  • #19
                    Originally posted by kcrudup View Post
                    I have a Dell XPS (13" 2-in-1) that's got the 15W i7-1065G7. I can get constant 45W performance (@95C) by using a utility called "Throttled" and disabling thermald. I use a utility called "S-Tui" to monitor temps, wattages and CPU Frequency (along with "turbostat"). Intel P-State (incl. HWP) and the "powersave" governor are unchanged from stock (the "throttled" utility modifies some MSR tables/registers).

                    I keep saying I'm going to build some of these same tests Michael runs and compare them against my setup (bleeding-edge kernel, mitigations off, custom thermal management) and see if I get better than the same middle-of-the-pack outcomes he gets with the stock configs, maybe now is the time to start.

                    What's nice is between a (only slightly-modified) "tlp" config and "throttled" I get 45W sustained on AC when needed, but idle wattages as low as .9W on battery.
                    Dell XPS with i7-8550U CPU here with tlp and run -100mV undervolt on cpu (stable with mprime and s-tui).
                    In the default config, this CPU will reach 3.4GHz all core @ 35W for a few seconds before reaching 80C and throttling to 1.8GHz base clock @ 15W.
                    With -100V, it will do 2.0-2.4GHz @ 15W and 80C all day long.
                    Throttled could probably improve this but I'm not particularly thrilled of running CPUs >80C (maybe I'm "old school").

                    Overall, pretty mediocre, especially considering the cost. Glad to see some healthy competition.

                    Comment


                    • #20
                      Originally posted by alcalde View Post
                      Numpy uses a lot of C extensions. Was it compiled from source or does the Python package manager just pull in the generic Linux binary blob they have? It could be that the default Numpy binary is optimized for Intel. Some python distributions have offered numpy with a BLAS library compiled with Intel's compiler, known to use no optimizations if it detects an AMD CPU.
                      Most Linux distros ship without optimized Blas and Lapack for compatibility. A good portion of scientific software tends to default to MKL. Proprietary software like Matlab is notorious for that but also Numpy may default to that depending on install. MKL tends to fall back to SSE on AMD but there are hacks
                      https://www.reddit.com/r/Amd/comment...d_and_the_mkl/

                      That being said, I'd be surprised if Michael would run numeric performance tests without optimized linear algebra libraries.
                      Last edited by mppix; 12 May 2020, 09:31 PM.

                      Comment

                      Working...
                      X