At some point this year the big buzzword appears to have shifted away from “the cloud” to “deep learning.” Eugen Weber once remarked that everybody who’s anybody sacked Rome. In similar fashion, everybody who’s anybody is setting machine learning loose on problems — especially the much-beloved process of deep learning. Deep learning is a machine learning method that extracts information from a volume of data by crunching through millions of entries, in order to sift out and rank the most important aspects of the data. One current wunderkind of deep learning is the Jeopardy-playing, cancer-diagnosing AI Watson, which runs on IBM’s Power architecture.
Now IBM and Nvidia just jointly released a set of deep learning software tools called PowerAI, designed to use the Power architecture. These tools are meant to make machine learning projects faster and tighten up neural net performance, and in part, to provide a way to extend Watson as our understanding of AI advances. PowerAI consists of a set of binary distributions of several custom-tuned neural nets (software) and some associated custom Nvidia GPUDL libraries compatible with machine-learning tasks. As a software suite, PowerAI is designed to run on a single one of IBM’s highest-end Power servers, the Power S822LC for High Performance Computing (HPC), but also to scale up from one to many many supercomputing clusters.
Readers may be familiar with Nvidia’s NVLink communications protocol; NVLink is a hardware bridge that can link CPUs to GPUs or GPUs to GPUs, at up to 80 GB/s of bandwidth. On IBM’s highest-end S822LC HPC servers, because of hardware acceleration using the IBM POWER8 chip in tandem with the NVLink hardware bridge between NV GPUs, PowerAI can run Alexnet with Caffe or BVLC at twice the speed of comparable x86 systems.
Beyond Watson, one of the outfits making use of Power tools (yeah, they named it so we’d have to say it like that, very punny, guys) is the Human Brain Project, which is using Power supercomputers to crunch the massively parallel data flow through the human connectome, and even to simulate and visualize that data in a way that isn’t limited by the size of the output pipe. Another is the city of Yachay, Ecuador. Yachay is a planned city that its organizers have dubbed the “city of knowledge,” and said organizers intend to use PowerAI for calculations related to climate projections, food genomics, and energy research. The entire project is designed to push residents from emphasizing what they have to emphasizing what they know; the intent is to make the city run on knowledge-based innovation.
Source: Extremetech
0 commentaires:
Enregistrer un commentaire