mtm16-gpu

SGE

Connect to one of following with your mtm account:

ssh freki.ms.mff.cuni.cz
ssh geri.ms.mff.cuni.cz
ssh ufallab.ms.mff.cuni.cz

Connect to random solX node with cluster command.

qstat
qstat -u '*' -q gpu.q
qsub -N job_name -V -cwd -b y -j y "commnad; another_command"
qdel

-V exports environment variables to the qsub job
for unbuffred output: python -u
to exclude Titan: qsub -l h=!titan-gpu or use 2 slots (somewhat complicated on SGE)

Run this on solX

virtualenv venv --system-site-packages
. venv/bin/activate
pip install theano

THEANO_FLAGS='device=cuda'

cp -r /net/me/merkur3/MTM/mtm006/lab .
cd lab/mlpTut
./mlp.py