DELTA AI
Designed for today’s Artificial Intelligence and Machine Learning research needs
DeltaAI is a new resource that targets the computational needs of Artificial Intelligence/Machine Learning (AI/ML) workloads. Designed from the ground up to meet the demands of the AI/ML research community, DeltaAI hardware will be available in fall 2024 along with interfaces, operational policies and allocation methods.
To keep up with research needs, DeltaAI is working with AI/ML projects to define and enable the new and existing access and usage modalities best suited to the AI/ML community, including evaluating web interfaces, interactive use, persistent services, node use and more. DeltaAI will build upon existing Delta infrastructure, such as Open OnDemand, Science Gateways,user configurable service nodes, and more.
Improving the allocation process for AI/ML projects is essential to deploying a robust, highly accessible system, so the DeltaAI team is working with the U.S. National Science Foundation, ACCESS and the AI/ML community to create a smooth process for onboarding.
Delta AI Offers:
- DeltaAI will provide
- ~2X the performance of Delta: 21 FP64 PF and 633 FP16 PF from 320 NVIDIA Grace Hopper GPUs with 96GB per GPU (384GB GPU mem per node).
- Supported by 14 PB of storage at up to 1TB/sec.
- Interconnected with a highly scalable fabric.
