English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Zendy Plus

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Zendy Tools

Zendy Open

Basic Linear Algebra Subprograms (BLAS) are a set of low level linear algebrakernels widely adopted by applications involved with the deep learning andscientific computing. The massive and economic computing power brought forth bythe emerging GPU architectures drives interest in implementation ofcompute-intensive level 3 BLAS on multi-GPU systems. In this paper, weinvestigate existing multi-GPU level 3 BLAS and present that 1) issues, such asthe improper load balancing, inefficient communication, insufficient GPU streamlevel concurrency and data caching, impede current implementations from fullyharnessing heterogeneous computing resources; 2) and the inter-GPUPeer-to-Peer(P2P) communication remains unexplored. We then present BLASX: ahighly optimized multi-GPU level-3 BLAS. We adopt the concepts ofalgorithms-by-tiles treating a matrix tile as the basic data unit andoperations on tiles as the basic task. Tasks are guided with a dynamicasynchronous runtime, which is cache and locality aware. The communication costunder BLASX becomes trivial as it perfectly overlaps communication andcomputation across multiple streams during asynchronous task progression. Italso takes the current tile cache scheme one step further by proposing aninnovative 2-level hierarchical tile cache, taking advantage of inter-GPU P2Pcommunication. As a result, linear speedup is observable with BLASX undermulti-GPU configurations; and the extensive benchmarks demonstrate that BLASXconsistently outperforms the related leading industrial and academic projectssuch as cuBLAS-XT, SuperMatrix, MAGMA and PaRSEC.

BLASX: A High Performance Level-3 BLAS Library for Heterogeneous Multi-GPU Computing