97 98 99 100 101 1,895 Replies Latest reply on Jun 14, 2018 1:36 PM by constantinx Go to original post
      • 1,470. Re: gcc segmentation faults on Ryzen / Linux
        raydude

        biotista wrote:

         

        very good points. thank you for sharing.

         

        hypothetically, if windows was indeed patched with software, then does the explanation of a linux kernal fix taking "months" to write make sense? is it plausible that it is easier/cheaper to RMA linux users instead of patch the linux kernel whereas a windows patch would be done by Microsoft and be quicker/cheaper? if MS could patch windows then wouldnt ubuntu know exactly how to patch linux and it would be done super fast and cheap?

        I agree with oldamdfan's comments above.

         

        I'd like to add that if AMD don't know exactly what the problem is: there is no guarantee that threadripper and epyc are unaffected in spite what they have claimed. I haven't heard of the issue on threadripper yet though so that's a good sign.

         

        There is no software fix for this issue. If there were, it would have been implemented ages ago.

         

        IMO they are very lucky. If this issue had come out on Windows as hard as it has on linux, AMD would have lost a lot of reputation among gamers.

         

        And there's still the sudden reset issue that needs to be addressed. (my intuition is screaming BIOS because I've never seen it on my Gigabyte, but others have seen it on other motherboards).

        • 1,471. Re: gcc segmentation faults on Ryzen / Linux
          msel

          The issue is reproducible in Windows.  If you have an affected CPU, install cygwin and start compiling stuff.  The kill-ryzen script, with a few minor tweaks, works just as well in cygwin on Windows as it does in Linux for the purposes of generating a build failure during gcc compilation.  It took a little longer to reproduce the issue under Windows, but it's still there.  Taxing affected CPUs in a particular way makes the result unreliable.

           

          Building mesa actually worked best for me as a method of reproducing the failure.  Much faster than gcc.

          1 of 1 people found this helpful
          • 1,472. Re: gcc segmentation faults on Ryzen / Linux
            garretsw

            What is the easiest way to run the test script using wsl on windows? I have wsl running but using wget http://funks.ddns.net:8080/tools/ryzen/testRyzenGCC.sh doesn't work for some reason

            • 1,473. Re: gcc segmentation faults on Ryzen / Linux
              shmerl

              msel wrote:

               

              Building mesa actually worked best for me as a method of reproducing the failure. Much faster than gcc.

               

              Yep, that's what I was using to test segfaults on Linux.

              • 1,474. Re: gcc segmentation faults on Ryzen / Linux
                ttbek

                I don't understand why anyone is subtracting anything....  You're right

                about ambient though. .. it shouldn't be even 1 degree below ambient.   35

                is a reasonable idle temp for a room temperature (as per standard 23

                degrees C) ambient.  I see about 37 in the bios myself, which I believe is

                correct.  I'm in Qatar but we have this wonder we call "AC," so ambient is

                quite reasonable indoors.  From lmsensors I see like 16 idle so clearly

                about 20 needs to be added to my lmsensors reading but I don't see why

                anyone should be subtracting from anything even once,  never mind twice.

                • 1,475. Re: gcc segmentation faults on Ryzen / Linux
                  shmerl

                  ttbek wrote:

                   

                  I don't understand why anyone is subtracting anything....

                  It's explicit feature of X series, that AMD announced. They report temperature with +20°C offset.

                  • 1,476. Re: gcc segmentation faults on Ryzen / Linux
                    ttbek

                    Ah, I see

                    https://community.amd.com/community/gaming/blog/2017/03/13/amd-ryzen-community-update

                    But it seems like all the MB manufacturers are accounting for it?  My CPU

                    is an 1800X and it wouldn't make any sense if my CPU was 17 degree C.

                    • 1,477. Re: gcc segmentation faults on Ryzen / Linux
                      scorpio810

                      garretsw a écrit:

                       

                      What is the easiest way to run the test script using wsl on windows? I have wsl running but using wget http://funks.ddns.net:8080/tools/ryzen/testRyzenGCC.sh doesn't work for some reason

                      What is your problem, you are just curious, or you 're a Linux software developer and use WSL, Cygwin, minGW under Windows?

                      • 1,478. Re: gcc segmentation faults on Ryzen / Linux
                        ttbek

                        http://www.alcpu.com/forums/viewtopic.php?f=64&t=5825&start=50

                        It appears the offset is enabled by default on the MSI Hero boards.

                        • 1,479. Re: gcc segmentation faults on Ryzen / Linux
                          bradc

                          ttbek wrote:

                           

                          Ah, I see

                          https://community.amd.com/community/gaming/blog/2017/03/13/amd-ryzen-community-update

                          But it seems like all the MB manufacturers are accounting for it? My CPU

                          is an 1800X and it wouldn't make any sense if my CPU was 17 degree C.

                          You do need to watch for it though. Early BIOS for the ASUS Prime X370 did not account for it, so I had to manually add an offset to my lm-sensors configuration. Without announcement that was rectified in a later BIOS and suddenly I was subtracting 20C from a setting the BIOS was now already subtracting 20C from (technically the BIOS programmed an offset into the IT SuperIO chip).

                           

                          So it is something to watch. Looking at the source for the updated k10temp driver, the offset is added based on the CPU name text, so that'll have to be kept up to date as new chips are released.

                           

                          I like Qatar. I spent quite a bit of time there when the new MOIA was being built.

                          • 1,480. Re: gcc segmentation faults on Ryzen / Linux
                            nop

                            shmerl wrote:

                             

                            ttbek wrote:

                             

                            I don't understand why anyone is subtracting anything....

                            It's explicit feature of X series, that AMD announced. They report temperature with +20°C offset.

                            Yes, I believe most of us are aware of that by now, but recent BIOSes of at least some mainboards already seem to correct that offset.

                            When I booted up my system the first time on the original BIOS I bought the mainboard (ASUS Prime X370-Pro) with, it also displayed very high temperatures which were obviously elevated by 20°C. After upgrading to the most recent BIOS available at this time the offset was gone and (hopefully) the real temperatures are shown now.

                            Since the temperatures you report are far from what seems to be possible, judged from decades of experience with that stuff and reports from people all over the world, I'm certain that your sensor readings are not reflecting the truth and were obviously "corrected" 2 times.

                            I'm also using a monstrous cooler with 2 large fans directly mounted on it (Noctua NH-D14) and I've been applying thermal paste for decades, not continuously but every now and then . All together there are 8 fans rotating in my tower case, so cooling should be more than appropriate.

                            • 1,481. Re: gcc segmentation faults on Ryzen / Linux
                              ttbek

                              I hope you're referring to the 17 C (obviously not correct) because 37 C is

                              very reasonable considering that the CPU is water cooled and my AC points

                              directly at my PC, about 2 meters away.  38 C (1 degree different) is even

                              used as the example real value in the post from AMD I referenced above when

                              I realized what they were talking about.  7 fans for my system, slightly

                              oversized PSU prevented an 8th.

                              • 1,482. Re: gcc segmentation faults on Ryzen / Linux
                                nop

                                37°C with good water cooling sound ok but temperatures below 30°C like the reported 26,5°C with air cooling or the 17°C which could no way be possible are not correct for sure.

                                I get values as low as 42°C when the CPU is idling, overclocked, with very good air cooling and ambient temperature at about 23°C.

                                ttbek wrote:

                                 

                                I hope you're referring to the 17 C (obviously not correct) because 37 C is

                                very reasonable considering that the CPU is water cooled and my AC points

                                directly at my PC, about 2 meters away. 38 C (1 degree different) is even

                                used as the example real value in the post from AMD I referenced above when

                                I realized what they were talking about. 7 fans for my system, slightly

                                oversized PSU prevented an 8th.

                                • 1,483. Re: gcc segmentation faults on Ryzen / Linux
                                  scorpio810

                                  nop a écrit:

                                   

                                  37°C with good water cooling sound ok but temperatures below 30°C like the reported 26,5°C with air cooling or the 17°C which could no way be possible are not correct for sure.

                                  I get values as low as 42°C when the CPU is idling, overclocked, with very good air cooling and ambient temperature at about 23°C.

                                   

                                  My CPU @stock need only 0.47V Vcore in idle....

                                  • 1,484. Re: gcc segmentation faults on Ryzen / Linux
                                    pjssilva

                                    Bad news from my front. I have an RMAed CPU of week 25. It sustained 3 days of kill-ryzen.sh using BIOS defaults. After that I updated my BIOS and changed my memory profile to use XMP 2 (2400 instead of 2100 from Bios default).  Unfortunately, under this conditions I got a new failure in one of the loops:

                                     

                                    -----------------

                                    /mnt/ramdisk/workdir/gcc-7.1.0/gcc/config/i386/i386.c: In function 'int standard_sse_constant_p(rtx, machine_mode)':

                                    /mnt/ramdisk/workdir/gcc-7.1.0/gcc/config/i386/i386.c:11699:8: internal compiler error: Segmentation fault

                                       mode = GET_MODE (x);

                                    0x11ac5ec crash_signal

                                        /mnt/ramdisk/workdir/gcc-7.1.0/gcc/toplev.c:337

                                    0x104058c create_output_operand

                                        /mnt/ramdisk/workdir/gcc-7.1.0/gcc/optabs.h:94

                                    -----------------

                                     

                                    I will now revert back memory to BIOS default. I think I read somewhere that Ryzen only officially supports 2100Hz when dealing with 64Gb.

                                     

                                    I have to admit that I am really tired of this. It will be hard to really trust this CPU after so many problems.

                                    1 of 1 people found this helpful
                                    97 98 99 100 101