/deep-learning-cpu-gpu-benchmark

Repository to benchmark the performance of Cloud CPUs vs. Cloud GPUs on TensorFlow, FloydHub, AWS and soon GCE.

Primary LanguageJupyter NotebookMIT LicenseMIT

deep-learning-cpu-gpu-benchmark

Repository to benchmark the performance of Cloud CPUs vs. Cloud GPUs on TensorFlow(as Keras Backend), FloydHub and AWS.

Usage

floydhub_keras_cpu_gpu_benchmark.py contains the launcher script for running the complete benchmark on FH. (You have to be logged on FloydHub before you run it)

aws_gce_keras_cpu_gpu_benchmark.py [aws|gce] contains the launcher script for running the complete benchmark on AWS or GCE. Note: you must specifies the cloud service name as a script parameters.*

/test_files contains the test files.

/logs contains the performance output (CSV) for each test.

/analysis contains the Jupyter Notebook of the logs used to create the interactive data visualizations.

TODO

See the project page: here.

Credits

Max Woolf (@minimaxir)

License

MIT