Addresses the most challenging deep learning issues
Deep Learning Impact is designed to address the deep learning lifecycle with a focus on the steps that are the most time consuming or require highly specialized knowledge—whether the iterative and time-consuming nature of the workflow, the lack of skills to train and tune models, the need to implement open source frameworks, the high demands for computing capacity or the challenges of scale.
Meet needs of high performance deep learning applications
This enterprise-grade solution supports multitenancy, elastic resource allocation, and a distributed training fabric designed to allow most applications to run in parallel without the need for code changes. Also supports training visualization and tuning, hyper-parameter search and optimization, and large model support, which leveraging CPU and GPU memory across a single large model.
Simplifies and optimizes an end-to-end workflow
This process spans installing and configuring the environment to ingestion of data; data preparation and transformation to meet the requirements of deep learning frameworks; building, training and optimizing the neural models that make deep learning possible; deploying the model in production; and improving the model by retraining using new data as needs evolve.
Takes advantage of a distributed server architecture
Deep Learning Impact enables data scientists to quickly ingest, transform, train and iterate by running the processes in parallel. Deep Learning Impact is built to take advantage of IBM Spectrum Conductor, a highly available multitenant framework designed to build a shared, enterprise-class Apache Spark environment, and provide centralized management and monitoring, along with end-to-end security.
- Distribution: Electronic download in multiple eAssemblies. Physical media not available.
- IBM PowerAI V1.5 base package, Red Hat Enterprise Linux 7.4 operating system
- Scalability: Up to 64 nodes with up to 256 GPUs
- IBM Power System S822LC for HPC (8335-GTB) servers