MKL용 AMI 만드는 법

Edit

MKL용 AMI 만드는 법

Marxico

먼저 Script로 MKL 설치

그리고 먼저 Nvidia Graphics Card를 깔아야한다.

g2.instance는 GRID 520 카드라서 이에 맞는 그래픽 카드를 설치한다.

그 전에 dpkg 사이즈가 커서 storage의 사이즈를 20GB 까지 늘린다.

Nvidia Graphics Card 설치하기

wget https://developer.nvidia.com/compute/cuda/8.0/Prod2/local_installers/cuda-repo-ubuntu1604-8-0-local-ga2_8.0.61-1_amd64-deb

$ wget http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1404/x86_64/cuda-repo-ubuntu1404_7.5-18_amd64.deb

$ sudo dpkg -i cuda-repo-ubuntu1404_7.5-18_amd64.deb

sudo apt-get update

sudo apt-get upgrade -y

sudo apt-get clean

쿠다를 설치한다.

sudo apt-get install -y cuda

sudo apt-get clean
CUDA 설치되어있는지 확인하기

nvidia-smi
Kernel Module이랑 device 설치되어있는지 확인하기

lsmod | grep -i nvidia

마지막으로 openblas를 설치한다

sudo apt-get install libopenblas-dev

그리고 마그마를 깐다.

wget http://icl.cs.utk.edu/projectsfiles/magma/downloads/magma-2.2.0.tar.gz

taf -zxvf magma…

cd magma

make.inc 중 make.inc.gcc-mkl 를 make.inc로 바꿔서 make 하기

cp make.inc-examples/make.inc.gcc-mkl .

mv make.inc.gcc-mkl make.inc

make -j16

그리고 환경변수를 설정한다.

vi ~/.bashrc

export CUDADIR=/usr/local/cuda
export PATH=

__SVG__b1890c3074560a992746043690abdb4c

CUDADIR/bin
export LD_LIBRARY_PATH=

__SVG__4c3b647260ff17f7179d0fed6e86b4aa

CUDADIR/lib64:/usr/local/magma/lib

source /opt/intel/bin/compilervars.sh intel64

source ~/.bashrc 하고

sudo make install prefix=/usr/local/magma를 한다

완료

%23%23%23%23%20MKL%uC6A9%20AMI%20%uB9CC%uB4DC%uB294%20%uBC95%20%0A@%28Marxico%29%0A%0A%uBA3C%uC800%20Script%uB85C%20MKL%20%uC124%uCE58%0A%0A%0A%3Cscript%20src%3D%22https%3A//gist.github.com/mjaysonnn/0135fb1161edaf647bc5126b4639c7f0.js%22%3E%3C/script%3E%0A%0A%uADF8%uB9AC%uACE0%20%20%uBA3C%uC800%20Nvidia%20Graphics%20Card%uB97C%20%uAE54%uC544%uC57C%uD55C%uB2E4.%0A%0Ag2.instance%uB294%20GRID%20520%20%uCE74%uB4DC%uB77C%uC11C%20%uC774%uC5D0%20%uB9DE%uB294%20%uADF8%uB798%uD53D%20%uCE74%uB4DC%uB97C%20%uC124%uCE58%uD55C%uB2E4.%0A%0A%uADF8%20%uC804%uC5D0%20dpkg%20%uC0AC%uC774%uC988%uAC00%20%uCEE4%uC11C%20storage%uC758%20%uC0AC%uC774%uC988%uB97C%2020GB%20%uAE4C%uC9C0%20%uB298%uB9B0%uB2E4.%0A%0ANvidia%20Graphics%20Card%20%uC124%uCE58%uD558%uAE30%0A%0Awget%20https%3A//developer.nvidia.com/compute/cuda/8.0/Prod2/local_installers/cuda-repo-ubuntu1604-8-0-local-ga2_8.0.61-1_amd64-deb%0A%0A%24%20wget%20http%3A//developer.download.nvidia.com/compute/cuda/repos/ubuntu1404/x86_64/cuda-repo-ubuntu1404_7.5-18_amd64.deb%0A%0A%20%24%20sudo%20dpkg%20-i%20cuda-repo-ubuntu1404_7.5-18_amd64.deb%0A%0Asudo%20apt-get%20update%0A%0Asudo%20apt-get%20upgrade%20-y%20%20%0A%0Asudo%20apt-get%20clean%0A%0A%uCFE0%uB2E4%uB97C%20%uC124%uCE58%uD55C%uB2E4.%0A%0Asudo%20apt-get%20install%20-y%20cuda%0A%0Asudo%20apt-get%20clean%0ACUDA%20%uC124%uCE58%uB418%uC5B4%uC788%uB294%uC9C0%20%uD655%uC778%uD558%uAE30%0A%0Anvidia-smi%0AKernel%20Module%uC774%uB791%20device%20%uC124%uCE58%uB418%uC5B4%uC788%uB294%uC9C0%20%uD655%uC778%uD558%uAE30%0A%0Alsmod%20%7C%20grep%20-i%20nvidia%0A%0A%0A%0A%uB9C8%uC9C0%uB9C9%uC73C%uB85C%20openblas%uB97C%20%uC124%uCE58%uD55C%uB2E4%0A%0Asudo%20apt-get%20install%20libopenblas-dev%0A%0A%0A%uADF8%uB9AC%uACE0%20%uB9C8%uADF8%uB9C8%uB97C%20%uAE50%uB2E4.%0A%0Awget%20http%3A//icl.cs.utk.edu/projectsfiles/magma/downloads/magma-2.2.0.tar.gz%0A%0Ataf%20-zxvf%20magma...%0A%0Acd%20magma%0A%0A%0Amake.inc%20%uC911%20make.inc.gcc-mkl%20%uB97C%20make.inc%uB85C%20%uBC14%uAFD4%uC11C%20make%20%uD558%uAE30%0A%0Acp%20make.inc-examples/make.inc.gcc-mkl%20.%20%0A%0Amv%20make.inc.gcc-mkl%20make.inc%0A%0A%0Amake%20-j16%20%0A%0A%0A%uADF8%uB9AC%uACE0%20%uD658%uACBD%uBCC0%uC218%uB97C%20%uC124%uC815%uD55C%uB2E4.%0A%0Avi%20%7E/.bashrc%0A%0Aexport%20CUDADIR%3D/usr/local/cuda%0Aexport%20PATH%3D%24PATH%3A%24CUDADIR/bin%0Aexport%20LD_LIBRARY_PATH%3D%24LD_LIBRARY_PATH%3A%24CUDADIR/lib64%3A/usr/local/magma/lib%0A%0Asource%20/opt/intel/bin/compilervars.sh%20intel64%0A%0Asource%20%7E/.bashrc%20%uD558%uACE0%0A%0A%0A%0Asudo%20make%20install%20prefix%3D/usr/local/magma%uB97C%20%uD55C%uB2E4%20%0A%0A%uC644%uB8CC%0A%0A%0A%0A%0A%0A%0A%0A%0A