Skip to content

4*. Setup GPU Server (Alternative)

JOHN THORPE edited this page Apr 2, 2021 · 1 revision

Chapter 4*. GPU Related Stuff

Previous page: 3. Prepare Input Dataset | Next page: 5. Run Dorylus | Home: Home

If GPUs are available on graph servers, an alternative way of computation using GPU is provided.

Required Dependencies

Since the GPU Server program resides on the same machine as graph server program.

Additional packages are required: cudatoolkit >=10.0 cuDnn>=7.6

In some cases, you may need to change Environment Variables

include .../cuda/bin #cuda bin location in $PATH include .../cuda-10.0/lib64 #cuda lib location in $LD_LIBRARY_PATH

Build&Run GPU Server

Use Graph$ ./gnnman/helpers/graphserver.bld gpu to build Graph server program with GPU support.

Use Graph$ ./run/run-onnode graph <Dataset> gpu # For GPU version to run GPU version.

Env Variables Setting

export PATH=$PATH:/usr/local/cuda/bin
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib64
export LG_RT_DIR="$HOME/legion/runtime"
export CUDA=/usr/local/cuda

Previous page: 3. Prepare Input Dataset | Next page: 5. Run Dorylus | Home: Home