GPUSetup

GPU Enablement

Build the GPU-enabled CLI tools.

1. cd aisp-core
2. mvn -DskipTests=true -P cuda10.2 clean install

This has created aisp-core/aisp-core-main/target/aisp-core-main-0.0.1-SNAPSHOT-cuda10.2.zip

On the target machine (With the GPUs)

Install the CLI tools

• cd SOMEDIR
• unzip aisp-core-main-0.0.1-SNAPSHOT-cuda10.2.zip
• export AISP_HOME=$SOMEDIR/aisp
• export PATH=$AISP_HOME/bin:$PATH
• sudo setup-aisp -gpu

Manually install Cuda10.2 and Cudnn (Linux)

Instead of using the setup-aisp tool, you can manually install the GPU support.

• Download and install Cuda10.2 (not 11.0) from the NVidia site at https://developer.nvidia.com/cuda-10.2-download-archive
• Download Cudnn v7 (not 8) from NVidia site at https://developer.nvidia.com/rdp/cudnn-archive. Unzip and copy contents of lib64 directory to cuda10.2 lib64 (/usr/local/cuda-10.2/lib64 on ubuntu).

Running the CLI

Before running the CLI tools you may want to set the memory options for DL4J. Something like the following

• export JAVA_OPTIONS="-Xmx32g -Dorg.bytedeco.javacpp.maxbytes=16g -Dorg.bytedeco.javacpp.maxphysicalbytes=48g"

DL4J has a page on memory if you have problems (in general maxphysicalbytes= maxbytes + java heap)

Test your setup using the train or evaluate CLI tools

evaluate -model dcase -sounds <your sounds> -label <your label>

If all goes well, you should see this in the log - Note that CUDA is in being used in the Backend.

...
[main] INFO org.nd4j.linalg.factory.Nd4jBackend - Loaded [JCublasBackend] backend
[main] INFO org.nd4j.nativeblas.NativeOpsHolder - Number of threads used for linear algebra: 32
[main] INFO org.nd4j.linalg.api.ops.executioner.DefaultOpExecutioner - Backend used: [CUDA]; OS: [Linux]
[main] INFO org.nd4j.linalg.api.ops.executioner.DefaultOpExecutioner - Cores: [8]; Memory: [32.0GB];
[main] INFO org.nd4j.linalg.api.ops.executioner.DefaultOpExecutioner - Blas vendor: [CUBLAS]
[main] INFO org.nd4j.linalg.jcublas.JCublasBackend - ND4J CUDA build version: 10.2.89
[main] INFO org.nd4j.linalg.jcublas.JCublasBackend - CUDA device 0: [Tesla V100-PCIE-16GB]; cc: [7.0]; Total memory: [16945512448]
...

I