BID

"The fastest Big Data tools on the Web.  BIDMat is an interactive matrix library that integrates CPU and GPU acceleration and novel computational kernels, and BIDMach is a machine learning system that includes very efficient model optimizers and mixing strategies.

BIDMat is a matrix library+interactive environment intended to support large-scale exploratory data analysis. It includes GPU and cluster computing support and is a sibling of BIDMach, a machine learning Toolkit. Together BIDMat and BIDMach on single nodes hold the performance records (including vs. cluster systems) for a large and growing list of machine learning problems. Here are our current benchmarks. Here is the project home page which includes links to compiled code bundles for both toolkits.

BIDMach is a very fast tool for machine learning, from small problems to terabyte scale. BIDMach is currently the fastest system for many common machine learning tasks (see the benchmarks section). In fact on a single GPU-equipped node, BIDMach outperforms the fastest cluster systems running on up to a few hundred nodes. BIDMach also scales well. BIDMach streams data off disk and is not memory-limited. With a large RAID, BIDMach has run topic models with hundreds of topics on several terabytes of data. We are aware of no other system able to solve that problem at comparable scale."

http://bid2.berkeley.edu/bid-data-project/

https://github.com/BIDData/BIDMat

https://github.com/BIDData/BIDMach

https://devblogs.nvidia.com/parallelforall/bidmach-machine-learning-limit-gpus/