Skip to content

Latest commit

 

History

History
38 lines (28 loc) · 1.35 KB

README.md

File metadata and controls

38 lines (28 loc) · 1.35 KB
Raw

mxnet_osx1011_mbp

A simple guide to install mxnet on a MBP using OSX 10.11

check your MBP's GPU

Make sure a NVIDIA discrete GPU card, e.g., GTX 750M, is on your MBP; w/o it, you only can expect running mxnet on CPU.

CUDA 7.5 and cuDNN V3

follow official instructions

compile mxnet C++

On OSX, need update config.mk as follows

  • using clang, not gcc
  • installing openblas through brew; using apple in the config.mk
  • based on this issue report, need setup
ADD_LDFLAGS += -Xlinker -F/Library/Frameworks -Xlinker -framework -Xlinker CUDA

Copy the provided config.mk file to the root-dir of the mxnet and then run make command

install mxnet-R package

  • follow this instruction to disable SIP (aka. rootless) on OSX 10.11
  • make sure R is > 3.2
  • have these path setups
# 12/2/2015
# for mxnet-R
export LD_LIBRARY_PATH=/usr/local/cuda/lib:$LD_LIBRARY_PATH
export DYLD_FALLBACK_LIBRARY_PATH=.:/usr/local/cuda/lib:$DYLD_FALLBACK_LIBRARY_PATH
  • using installation from the source
  • for RStudio, please start it from terminal so that the mxnet package can be loaded

Using the provided lenet.R to test running on both CPU and GPU