GPU accelerating your computation in Python

There are many powerful libraries in the Python ecosystem for accelerating the computation of large arrays with GPUs. We have CuPy for GPU array computation, Dask for distributed computation, cuML for machine learning, Pytorch for deep learning and more. We will dig into how these libraries can be used together to accelerate geoscience workflows and how we are working with projects like Xarray to integrate these libraries with domain-specific tooling. Sgkit is already providing this for the field of genetics and we are excited to be working with community groups like Pangeo to bring this kind of tooling to the geosciences.

How to cite: Tomlinson, J.: Distributing your GPU array computation in Python, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-7610, https://doi.org/10.5194/egusphere-egu22-7610, 2022.

Transcript

1. Jacob Tomlinson
   Senior Software Engineer, RAPIDS
   Dask Core Maintainer
   GPU accelerating your computation in Python
   EGU General Assembly 2022
   EGU22-7610, https://doi.org/10.5194/egusphere-egu22-7610, 2022.
   

   View Slide

2. 2
   RAPIDS
   https://github.com/rapidsai
   

   View Slide

3. 3
   Jake VanderPlas - PyCon 2017
   

   View Slide

4. 4
   Pandas
   Analytics
   CPU Memory
   Data Preparation Visualization
   Model Training
   Scikit-Learn
   Machine Learning
   NetworkX
   Graph Analytics
   PyTorch,
   TensorFlow, MxNet
   Deep Learning
   Matplotlib
   Visualization
   Dask
   Open Source Data Science Ecosystem
   Familiar Python APIs
   

   View Slide

5. 5
   cuDF cuIO
   Analytics
   GPU Memory
   Data Preparation Visualization
   Model Training
   cuML
   Machine Learning
   cuGraph
   Graph Analytics
   PyTorch,
   TensorFlow, MxNet
   Deep Learning
   cuxfilter, pyViz,
   plotly
   Visualization
   Dask
   RAPIDS
   End-to-End Accelerated GPU Data Science
   

   View Slide

6. 6
   RAPIDS Matches Common Python APIs
   CPU-based Clustering
   from sklearn.datasets import make_moons
   import pandas
   X, y = make_moons(n_samples=int(1e2),
   noise=0.05, random_state=0)
   X = pandas.DataFrame({'fea%d'%i: X[:, i]
   for i in range(X.shape[1])})
   from sklearn.cluster import DBSCAN
   dbscan = DBSCAN(eps = 0.3, min_samples = 5)
   y_hat = dbscan.fit_predict(X)
   

   View Slide

7. 7
   from sklearn.datasets import make_moons
   import cudf
   X, y = make_moons(n_samples=int(1e2),
   noise=0.05, random_state=0)
   X = cudf.DataFrame({'fea%d'%i: X[:, i]
   for i in range(X.shape[1])})
   from cuml import DBSCAN
   dbscan = DBSCAN(eps = 0.3, min_samples = 5)
   y_hat = dbscan.fit_predict(X)
   RAPIDS Matches Common Python APIs
   GPU-accelerated Clustering
   

   View Slide

8. 8
   Benchmarks: Single-GPU cuML vs Scikit-learn
   1x V100 vs. 2x 20 Core CPUs (DGX-1, RAPIDS 0.15)
   

   View Slide

9. 9
   Exactly as it sounds—our goal is to make
   RAPIDS as usable and performant as
   possible wherever science is done. We
   will continue to work with more open
   source projects to further democratize
   acceleration and efficiency in science.
   RAPIDS Everywhere
   The Next Phase of RAPIDS
   

   View Slide

10. 10
    Statistical genetics toolkit in Python
    

    View Slide

11. 11
    Integrations, feedback, documentation support, pull requests, new issues, or code donations welcomed!
    APACHE ARROW
    GPU OPEN ANALYTICS
    INITIATIVE
    https://arrow.apache.org/
    @ApacheArrow
    http://gpuopenanalytics.com/
    @GPUOAI
    RAPIDS
    https://rapids.ai
    @RAPIDSai
    DASK
    https://dask.org
    @Dask_dev
    Work with us
    Everyone Can Help!
    

    View Slide

12. THANK YOU
    Jacob Tomlinson
    jtomlinson@nvidia.com
    @_jacobtomlinson
    

    View Slide