Announcement

Collapse
No announcement yet.

Some Ryzen Linux Users Are Facing Issues With Heavy Compilation Loads

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

  • #11
    Originally posted by chuckula View Post

    The posts seem to indicate that the crashes were non-deterministic. I know it's not practical for long-term use, but did you try just running 1 GCC process to see if that avoids the crashes?
    I am one of the few that doesn't have an issue ( my results are 1st in the sheet, one of the perks of generating the form )

    I would agree there does not seem to be any commonality but equally there were a number of questionable setups: binutils using an older version (not switched), RAM was incorrectly timed (this is a common issue with alot of ryzen setups if the BIOS is just used to autoset - I activated XMP to get RAM timings)

    GCC has been very suspect:
    https://wiki.gentoo.org/wiki/Ryzen

    Haswell for GCC-5.x has been shown to be bad
    GCC-6.x (1st with znver1) is non-optimised (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80313)
    GCC-7.1 so far is quite nice & a couple of us are using 7.1



    Comment


    • #12
      I posted in that Gentoo thread just before seeing this article. My finding is that -march=native with GCC 6.3.0 is bad. I hadn't rebuilt many packages yet but I had built my kernel with that and I was getting random freezes. Since switching to a generic build, it's been up for nearly 14 hours now. I'm about to try -march=bdver4 -mno-fma4 -mno-tbm -mno-xop -mno-lwp.
      Last edited by Chewi; 06-02-2017, 05:10 PM.

      Comment


      • #13
        I'm not seeing those errors currently, but I have seen them in the past via an over-aggressive overclock, or an overclock with insufficient voltage. Kernel compiles will do it, but mesa seems to cause it much more quickly.

        My 1800x, overclocked to 4.0ghz @1.35v run rock solid compiles all day long when properly cooled. Drop that voltage or increase the clock to 4.1Ghz, and I'll see segfaults as well as system crashes.

        Comment


        • #14
          I'm going to back the unstable overclocks theory. I had been overclocking my R7 1700 to 3.95 GHz, but after an hour or so, it would crash while doing compile jobs. Now I am running at 3.9 GHz and it runs all day everyday no problem, no instability.

          Comment


          • #15
            I too have experienced this issue.

            I've got a Crosshair VI Hero and Ryzen 1700X (all stock BIOS settings) and Fedora 25 with Linux 4.11.

            I've seen random segfaults compiling Linux with GCC, and random compilation failures (compilation would abort with a "compiler internal error", IIRC) compiling Chromium with Clang. Definitely doesn't seem GCC-specific to me.

            I've also run 48-hour Prime95 sessions with no errors on this box.

            Now that this has gotten more press, I should retest some more.

            Comment


            • #16
              Interesting.
              At work, this week we got notified of PC upgrades(we are software developers) and all were with Intel processors, so I recommended AMD processors because they are cheaper and have more cores(good for parallel compilation).
              The IT guys came with a negative response because currently, entire desktops with AMD processors are only for gaming setups(they do not buy parts to assemble them later).
              Good! We use GCC exclusively... That would of been a bad decision...

              Comment


              • #17
                Originally posted by chuckula View Post
                Have they maybe been able to narrow down the platforms that are or aren't affected? RyZen's launch was plagued by quite a bit of flakiness because the chips came out before the platform was 100% baked. Maybe some motherboards are OK and others are having problems.

                Additionally -- and I have to say this because it happens -- please don't tell me these guys are running some overclocked-to-the-edge setup and are getting shocked by segfaults.
                =) This. I do have Ryzen 1800X and samsung b-die, running the memory at 3200mhz cl 14- 15 1T, is no problems for me. Running it higher at 3466mhz or 3600mhz, could lead to some segfaults when compiling. Many mem sticks won't go as high as 3200 still, so there could be lot's of people running systems that seem stable enough for every day use, but fail at heavy load compile. But it will pass 48 threads of mprime for an hour. This is the case for me atleast, i downclocked my memory to 3200 cos it failed to compile WINE. There could of course be a bug in the software/microcode level too, it's all still evolving.

                Comment


                • #18
                  Originally posted by LinuxID10T View Post
                  I had been overclocking my R7 1700 to 3.95 GHz, but after an hour or so, it would crash while doing compile jobs. Now I am running at 3.9 GHz and it runs all day everyday no problem, no instability.
                  50/3900 = +1.28% speed increase and all you have to endure is a crash every hour? where do I have to sign... /s

                  Honestly, why do people even bother with these overclocks. If the chips were stable at a higher clock speed, they would sell them as higher speed chips.

                  Comment


                  • #19
                    Originally posted by jarfil View Post

                    50/3900 = +1.28% speed increase and all you have to endure is a crash every hour? where do I have to sign... /s

                    Honestly, why do people even bother with these overclocks. If the chips were stable at a higher clock speed, they would sell them as higher speed chips.
                    Well but then again, when u have it, why not push it to the limit? =)

                    ## cpu 4.0ghz 1.395v, mem 1.395v 3466mhz 14-14-14-14-38 T1
                    https://browser.geekbench.com/v4/cpu/2915025

                    Getting it stable with higher oc is getting easier with each bios update.

                    Comment


                    • #20
                      I had this issue on ASUS AM1M-A with AMD Athlon(tm) 5150 APU with Radeon(tm) R3

                      turned out to be a ram or ram combination issue (if I underclock the ram on default it works ok, on built in ram overclock profile it gets unstable under severe load)

                      Comment

                      Working...
                      X