AMD ROCm Installation Notes

rocm_install_notes

sudo apt-get update
sudo apt-get upgrade
sudo reboot

**** NOTE: did not install 5.0 kernel image; did not run dist-upgrade

sudo gedit /etc/default/grub
--> change the following
 GRUB_TIMEOUT=1
 GRUB_CMDLINE_LINUX_DEFAULT=""
 GRUB_SAVEDEFAULT=true
 GRUB_DEFAULT=saved

sudo update-grub

sudo apt install libnuma-dev glibc-doc glibc-doc-reference libc-dev-bin
wget http://repo.radeon.com/rocm/apt/debian/rocm.gpg.key
sha1sum rocm.gpg.key
echo 'deb [arch=amd64] http://repo.radeon.com/rocm/apt/debian/ xenial main' | sudo tee /etc/apt/sources.list.d/rocm.list
sudo reboot


**** Note: only rocm-dkms is strictly required
sudo apt install rocm-dkms cmake autoconf automake libtool flex bison gcc-doc python-doc python3-doc python3-nose python3-nose gnu-standards autoconf-doc bison-doc cmake-doc flex-doc python-nose-doc python3-examples

echo 'export PATH=$PATH:/opt/rocm/bin:/opt/rocm/profiler/bin:/opt/rocm/opencl/bin/x86_64' | sudo tee -a /etc/profile.d/rocm.sh
sudo usermod -a -G video $LOGNAME 

sudo reboot


sudo apt-get install build-essential clang clang-format clang-tidy cmake cmake-qt-gui ssh curl apt-utils pkg-config g++-multilib git libunwind-dev libfftw3-dev libelf-dev libncurses5-dev libpthread-stubs0-dev vim gfortran libboost-program-options-dev libssl-dev libboost-dev libboost-system-dev libboost-filesystem-dev rpm wget libboost1.65-doc libboost-atomic1.65-dev libboost-numpy1.65-dev

sudo apt-get install rocm-dkms rocm-dev rocm-libs hipcub rccl rocm-device-libs hsa-ext-rocr-dev hsakmt-roct-dev hsa-rocr-dev rocm-opencl rocm-opencl-dev rocm-utils miopen-hip miopengemm rocminfo rocm-profiler rocm_bandwidth_test rocr_debug_agent

sudo reboot


rocm_agent_enumerator
  gfx000
  gfx803

rocm_bandwidth_test
  Unidirectional copy peak bandwidth GB/s
  D/D       0           1           
  0         N/A         11.257026   
  1         11.117093   32.398932   

  Bdirectional copy peak bandwidth GB/s
  D/D       0           1           
  0         N/A         14.794945   
  1         14.794945   N/A  

dmesg | grep kfd
  [    3.328899] kfd kfd: Allocated 3969056 bytes on gart
  [    3.329348] kfd kfd: added device 1002:67df




sudo apt-get install python-numpy python-dev python-wheel python-mock python-future python-pip python-yaml python-setuptools python-crypto-doc python-cryptography-doc python-cryptography-vectors python-numpy-doc python-nose

sudo apt-get install python3-numpy python3-dev python3-wheel python3-mock python3-future python3-pip python3-yaml python3-setuptools

sudo pip install tensorflow-rocm --upgrade

python
  >>> import tensorflow
  >>> tensorflow.__version__
  '1.14.2'
  >>>


wget https://codeload.github.com/tensorflow/benchmarks/zip/abb1aec2f2db4ba73fac2e1359227aef59b10258
unzip benchmarks-master.zip

cd ~/benchmarks/scripts/tf_cnn_benchmarks


python tf_cnn_benchmarks.py --num_gpus=1 --batch_size=64 --model=resnet50
  [...]
  MIOpen(HIP): Warning [FindRecordUnsafe] File is unreadable: /opt/rocm/miopen/share/miopen/db/gfx803_32.cd.pdb.txt
  MIOpen(HIP): Warning [FindRecordUnsafe] File is unreadable: /opt/rocm/miopen/share/miopen/db/gfx803_32.cd.pdb.txt
  [...]
  Step  Img/sec                                 total_loss
  1     images/sec: 52.2 +/- 0.0 (jitter = 0.0) 8.220
  10    images/sec: 52.1 +/- 0.0 (jitter = 0.1) 7.880
  20    images/sec: 52.0 +/- 0.0 (jitter = 0.1) 7.910
  30    images/sec: 52.0 +/- 0.0 (jitter = 0.1) 7.820
  40    images/sec: 52.0 +/- 0.0 (jitter = 0.1) 8.005
  50    images/sec: 52.0 +/- 0.0 (jitter = 0.1) 7.768
  60    images/sec: 51.9 +/- 0.0 (jitter = 0.1) 8.114
  70    images/sec: 51.9 +/- 0.0 (jitter = 0.1) 7.817
  80    images/sec: 51.9 +/- 0.0 (jitter = 0.1) 7.978
  90    images/sec: 51.9 +/- 0.0 (jitter = 0.1) 8.099
  100   images/sec: 51.9 +/- 0.0 (jitter = 0.1) 8.042
  ----------------------------------------------------------------
  total images/sec: 51.87
  ----------------------------------------------------------------


sudo cp /opt/rocm/miopen/share/miopen/db/gfx803_36.cd.pdb.txt /opt/rocm/miopen/share/miopen/db/gfx803_32.cd.pdb.txt
python tf_cnn_benchmarks.py --num_gpus=1 --batch_size=64 --model=resnet50
  [...]
  Step  Img/sec                                 total_loss
  1     images/sec: 52.5 +/- 0.0 (jitter = 0.0) 8.220
  10    images/sec: 52.6 +/- 0.0 (jitter = 0.1) 7.880
  20    images/sec: 52.5 +/- 0.0 (jitter = 0.1) 7.910
  30    images/sec: 52.5 +/- 0.0 (jitter = 0.1) 7.821
  40    images/sec: 52.5 +/- 0.0 (jitter = 0.1) 8.004
  50    images/sec: 52.5 +/- 0.0 (jitter = 0.1) 7.769
  60    images/sec: 52.5 +/- 0.0 (jitter = 0.1) 8.114
  70    images/sec: 52.4 +/- 0.0 (jitter = 0.1) 7.819
  80    images/sec: 52.4 +/- 0.0 (jitter = 0.1) 7.979
  90    images/sec: 52.4 +/- 0.0 (jitter = 0.1) 8.096
  100   images/sec: 52.4 +/- 0.0 (jitter = 0.1) 8.026
  ----------------------------------------------------------------
  total images/sec: 52.38
  ----------------------------------------------------------------



rocm-smi
  ========================ROCm System Management Interface========================
  ================================================================================
  GPU  Temp   AvgPwr    SCLK     MCLK     Fan     Perf  PwrCap  VRAM%  GPU%
  0    71.0c  120.002W  1233Mhz  1750Mhz  38.82%  high  120.0W   97%   100%
  ================================================================================
  ==============================End of ROCm SMI Log ==============================



**** NCF and densenet require tensorflow models to be installed:
wget https://codeload.github.com/tensorflow/models/zip/v1.13.0
unzip v1.13.0
sudo mv models-1.13.0 /usr/local/share/
sudo chown -R root:root /usr/local/share/models-1.13.0
export PYTHONPATH=$PYTHONPATH:/usr/local/share/models-1.13.0/
my setup ==
  export PYTHONPATH=/usr/local/share/models-1.13.0