IBM’s data storage and AI portfolio now support the recently announced Nvidia DGX A100, dedicated to analytics and AI workloads.
David Wolford, marketing manager for IBM’s global cloud storage portfolio, wrote in a company blog last week: ” IBM has combined its file and object storage infrastructure with the Nvidia DGX A100 to create an end-to-end end solution. It has built-in capabilities that integrate data from IBM Cloud Object Storage and IBM Spectrum Scale storage of all data from Nvidia AI solutions for cataloging and discovery (in real-time)”.
Big Blue positions IBM Storage for Data and AI as three-tier AI components in the project pipeline; acquisition, transformation, and analysis/training. There are five products.
Cloud Object Storage (COS) Data Lake Storage
Spectrum Discover Software for file classification and indexing
Spectrum Scale horizontal scaling software for parallel file access.
ESS 3000 – NVMe drive array with full flash memory and containerized installation under Linux Spectrum Scale software with 24 SSD bays in a 2U cabinet.
Spectrum LSF (Load Sharing Tool) – Workload management and policy-driven operations for the high-performance computing planning system
IBM’s view on the storage and AI project pipeline
IBM is updating Nvidia’s draft solution to provide support for the DGX-A100. The new servers use the Tesla DGX A100 GPU, and Nvidia says the AI works better than the previous DGX-2. The Tesla V100 used in the Tesla V100 is 20 times faster.
The IBM Blueprint recommends that COS store the recorded data and act as a data lake. the Spectrum Discover becomes. indexes this data and adds metadata tags to its files. The LSF manages the AI project workflow through the Spectrum Discover triggers that transfer selected data from COS to the ES3000. where it powers the GPU in the A100 while the AI model is being developed and trained.
Dell, Igneous, NetApp, Pure Storage and VAST data are among the storage vendors that will also support the DGX A100. Some might try to cover the AI pipeline with a single storage array.
Copyright notice: This article is reprinted for the purpose of transmitting more information. If the source is incorrectly marked or infringes your legal rights, please contact us here, we will correct and delete it in time, thank you for your support and understanding.