Ginkgo is a production-ready, sparse linear algebra library used for HPC on GPU architectures. It’s now using oneAPI cross-architecture programming to support its foundational design with a high level of performance portability, and focus on software sustainability. ExpertsHartwig Anzt at Karlsruhe Institute of Technology (KIT) and Univ. of Tennessee, and Terry Cojean of KIT provide their insights on lessons learned moving CUDA code to other hardware architectures, and tools that help smooth that transition. “…The oneAPI ecosystem has proven to be a very powerful and useful option for us to actually target different architectures that are all supported by oneAPI…” Listen in. [21:19]

Guests:
Hartwig Anzt, research scientist at University of Tennessee, and group leader at Karlsruhe Institute of Technology (KIT)
Terry Cojean, lead developer of Ginkgo software at KIT
George Silva, Intel academic programs – program manager

To learn more:
Ginkgo project (GitHub)
Ginkgo project using oneAPI (Intel DevMesh)
Intel oneAPI Toolkits | Intel DPC++ Compatibility Tool| Free Download of Intel oneAPI Base Toolkit
Intel DevMesh for oneAPI projects

Read/Download the transcript

Tags: C++, developers, DPC++, George Silva, Ginkgo, Ginkgo project, Hartwig Anzt, heterogeneous programming, Intel, Intel DPC++ Compatibility Tool, Intel oneAPI Toolkit, oneAPI, sparse linear library, Terry Cojean
 
Posted in: Audio Podcast, Code Together, Intel