UXL Foundation

Don't miss the next two Special Interest Group (SIG) meetings happening this week! Check out the calendar on GitHub and join the community and developers working on implementations of the #oneAPI specification. https://bit.ly/3JMT1Xm #UXLFoundation

We are delighted to announce that the Università degli Studi di Salerno, Tsavorite Scalable Intelligence, and Dudaji Vietnam joined the #UXLFoundation as Contributing Members. Please join us in welcoming them! https://bit.ly/4ewhm1z #oneAPI

Today, we are excited to announce that CDACINDIA has joined the #UXLFoundation as one of our General Members. Welcome! https://bit.ly/4hIMDzo #oneAPI

The #UXLFoundation advances with the latest #oneAPI Specification 1.4 release, enhancing cross-platform development for CPUs, GPUs, FPGAs, and AI accelerators. Key updates include oneMath's (formerly oneMKL) revamped mathematical APIs, improved type safety, and new oneDPL algorithms, empowering developers to achieve high performance and portability across diverse hardware. Learn more: https://lnkd.in/gf23NcF9 #AI #OpenSource

New year, new #UXLFoundation Special Interest Group (SIG) meetings! Check out the calendar of upcoming meetings on GitHub and join the community and developers working on implementations of the #oneAPI specification. https://bit.ly/3JMT1Xm

Introducing GPU-BLOB: a benchmark to evaluate GPU offload thresholds for GEMM and GEMV kernels on modern HPC systems. Results reveal how problem shape, data reuse, and transfer methods impact performance, guiding optimal CPU-GPU utilization. Learn more: https://bit.ly/41KqAUh

The Barnes-Hut algorithm revolutionizes N-body simulations, cutting complexity from O(N²) to O(N log N). This study presents two portable parallel implementations using ISO C++: an unbalanced octree and a Hilbert-sorted BVH. Results show competitive CPU and GPU performance, with the par execution policy excelling in handling irregular algorithms on diverse hardware. Learn more: https://bit.ly/3Dnb5rc

Introducing the Tree-Based Model Divergence (TBMD) metric: a novel way to evaluate HPC programming productivity. Tested with the SilverVale framework, TBMD highlights trade-offs in performance and productivity across models like OpenMP, CUDA, and SYCL. https://bit.ly/4fsuUuM

In a benchmarking study, Intel's Data Center GPU Max 1550 (PVC) proves its mettle, showing competitive performance against NVIDIA H100 and AMD MI250 GPUs. Key highlights include linear scalability with OpenMP, SYCL, and MPI, robust real-world application results, and insights into system configurations. PVC emerges as a strong contender for high-performance computing. Learn more: https://bit.ly/4030dHZ

The UXL Foundation is leading the future of computing with open standards and multi-vendor collaboration. From AI to autonomous vehicles, our projects accelerate innovation across CPUs, GPUs, and specialized processors. In 2024, we expanded our community, launched impactful work, and set the stage for the next era of heterogeneous computing. Learn more from our annual report: https://bit.ly/3ZPva0U #UXLFoundation #oneAPI #OpenSource