多卡自检不通过 paddlepaddle-gpu-1.8.4.post107
Created by: yulinxuanzhi
系统:Ubuntu 16.04 GPU:4卡 P100 cudatoolkit=10.0.130 cudnn=7.6.5
自检报错信息: Your Paddle Fluid works well on SINGLE GPU or CPU. W0902 10:14:18.053751 36696 init.cc:226] Warning: PaddlePaddle catches a failure signal, it may not work properly W0902 10:14:18.053872 36696 init.cc:228] You could check whether you killed PaddlePaddle thread/process accidentally or report the case to PaddlePaddle W0902 10:14:18.053881 36696 init.cc:231] The detail failure signal is:
W0902 10:14:18.053891 36696 init.cc:234] *** Aborted at 1599012858 (unix time) try "date -d @1599012858" if you are using GNU date *** W0902 10:14:18.057443 36696 init.cc:234] PC: @ 0x0 (unknown) W0902 10:14:18.057744 36696 init.cc:234] *** SIGSEGV (@0x0) received by PID 36696 (TID 0x7f518d16b700) from PID 0; stack trace: *** W0902 10:14:18.060323 36696 init.cc:234] @ 0x7f518cd56390 (unknown) W0902 10:14:18.062988 36696 init.cc:234] @ 0x0 (unknown) Segmentation fault (core dumped)