Search papers, labs, and topics across Lattice.
University of Stuttgart
5
0
5
Agentic AI can generate parallel Julia code, but struggles with scalability and robustness in high-performance computing tasks.
Asynchronous tasks in HPX can deliver up to 26% faster performance than OpenMP in tiled Cholesky decomposition, challenging the dominance of traditional fork-join models.
Despite RISC-V's promise, this study reveals it's currently 14x slower than x86-64 for single-core Gaussian Process tasks, signaling a need for architectural improvements before it can compete in ML.
Asynchronous many-task runtimes like HPX can outperform MPI in computation-heavy radiation hydrodynamics simulations, but scalability bottlenecks remain.
GPU-acceleration of Gaussian Process regression, using HPX for asynchronous task management, beats cuSOLVER's performance by up to 11% for large datasets.
