What type of study is this?

Efficient GPU-Centered Singular Value Decomposition Using the Divide-and-Conquer Method

View Full Paper

Ask AI

Bookmark

Authors

New GPU-based SVD algorithm boosts performance in matrix factorization, eliminating CPU-GPU data transfers.

The GPU-centered SVD algorithm achieves speedups of up to 1293.64x compared to existing methods.
A novel divide-and-conquer approach reduces the need for CPU-GPU data transfers, enhancing efficiency.
Integrating related computations optimizes BLAS utilization and increases arithmetic intensity.
Experimental results validate that the new method significantly outperforms rocSOLVER/cuSOLVER and MAGMA.

Liu et al. (2025) studied this question.

View Full Paper

Ask AI

Bookmark