Loris Bennett <[email protected]> writes: > Hi, > > Still trying to compile TensorFlow, I was running into the problem that > the local drive, being only 14 GB, was not big enough for the temporary > build directories. > > Following the tip here: > > https://easybuild.readthedocs.io/en/latest/Configuration.html#buildpath > > I set /dev/shm for the build path as I have 192 GB RAM: > > eb TensorFlow-1.15.0-fosscuda-2019b-Python-3.7.4.eb --robot > --cuda-compute-capabilities=6.1,7.5 --buildpath=/dev/shm > > However, the build still fails, because stuff is still being written to /tmp: > > Wed Feb 26 15:13:02 CET 2020 : === Preparing sources in dir: > /tmp/eb-lffQRO/tmp.cYgY4xLGrb > > /dev/shm/TensorFlow/1.15.0/fosscuda-2019b-Python-3.7.4/TensorFlow/tensorflow-1.15.0 > > /dev/shm/TensorFlow/1.15.0/fosscuda-2019b-Python-3.7.4/TensorFlow/tensorflow-1.15.0 > cp: cannot create regular file > '/tmp/eb-lffQRO/tmp.cYgY4xLGrb/tensorflow/include/external/./aws/aws-cpp-sdk-core/include/aws/core/Aws.h': > No space left on device > > Is there some other variable I need to be setting?
Pointing --tmpdir to our large scratch file system did the trick. Cheers, Loris -- Dr. Loris Bennett (Mr.) ZEDAT, Freie Universität Berlin Email [email protected]
