World's Fastest Machine Learning With GPUs
2 likes389 views
The document discusses H2O4GPU, an open-source machine learning library that leverages GPU computing to significantly enhance performance, outperforming traditional CPU-based implementations. Key features include its compatibility with popular libraries like scikit-learn and its ability to accelerate various algorithms, including k-means and gradient boosting machines. The library demonstrates substantial speed improvements, up to 50 times for k-means compared to existing GPU implementations.
World's Fastest Machine Learning With GPUs
- 1. World's Fastest Machine Learning With GPUs http://github.com/h2oai/h2o4gpu Speaker: Jonathan C. McKinney
- 2. H2O4GPU TEAM Mateusz Erin Navdeep Rory Terry Karen Arno Jonathan Steve
- 5. 5 RISE OF GPU COMPUTING GPU-Computing perf 1.5X per year 1000X by 2025 102 103 104 105 106 107 Single-threaded perf 1.5X per year 1.1X per year APPLICATIONS SYSTEMS ALGORITHMS CUDA ARCHITECTURE
- 6. H2O4GPU / Open-Source: http://github.com/h2oai/h2o4gpu / Used within our own Driverless AI Product to boost performance 30X / Scikit-Learn Python API (and now R API) / All Scikit-Learn algorithms included / Important algorithms ported to GPU
- 10. K-Means • Significantly faster than Scikit-learn implementation (50x) • Significantly faster than other GPU implementations (5x-10x) • Supports kmeans++/kmeans|| initialization • Supports multiple GPUs • Supports batching data if exceeds GPU memory
- 12. 12 10 with latest solver
- 14. Principle Component Analysis (PCA)
- 16. Gradient Boosting Machines / Based upon XGBoost / Raw floating point data -> Binned into Quantiles / Quantiles are stored as compressed instead of floats / Compressed Quantiles are efficiently transferred to GPU / Sparsity is handled directly with highly GPU efficiency / Multi-GPU by sharding rows using NVIDIA NCCL AllReduce