
AWS FSx for Lustre Bolsters HPC and AI Capabilities with Support for EFA and NVIDIA GPU Direct Storage
$1,500.00
Authors: Mark Nossokoff and Jaclyn Ludema
Publication Date: December 202024
Length: 1 pages
At its recent re:Invent conference in Las Vegas, AWS announced its AWS FSx for Lustre service now supports its Elastic Fabric Adapter (EFA) high performance network interface and NVIDIA’s GPUDirect Storage (GDS). These new capabilities aim to reduce overall workload costs and substantially improve workload completion times by reducing inter-node latency for distributed training, more efficiently scaling across multiple GPUs, and eliminating CPU bottlenecks by creating a direct path between GPU and storage and reducing memory copying operations.
Related Products
AI Engineers in India Alleviate Effects of Water Scarcity
Tom Sorensen, Alex Norton
The August 2021 issue of the International Research Journal of Engineering and Technology (IRJET), a peer-reviewed research journal, included a paper based on the work of three researchers from India's St. Francis Institute of Technology (SFIT) summarizing their use of artificial intelligence (AI) and machine learning (ML) methods to help alleviate water shortages in India caused by population growth, urbanization, and climate change. Verlekar, Shah, and Kulkarni used a machine learning model to create a proactive scheme for managing local water resources, work that was prompted by a 2019 drought that impacted the Chennai area of India.
October 202021 | HYP_Link
New Error Correction Scheme Seeks to Advance Quantum Computing Capabilities
Bob Sorensen, Tom Sorensen
Researchers at the US-based Lawrence Berkeley National Lab (LBNL) recently reported a new approach to error mitigation in a quantum computer (QC) that targets error-producing noise, a ubiquitous problem that can severely limit the performance and utility of existing and near-future quantum computers. The method developed at LBNL consists of taking an initial noisy target circuit and constructing an analogous estimation circuit that is configured specifically for accurate noise characterization. The information gathered from running the estimation circuit is then applied to correct the noise in the original target circuit.
3 202022 | HYP_Link