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/

Global: Zendy Plus annual

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

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Along with the fast evolution of deep neural networks, the hardware system isalso developing rapidly. As a promising solution achieving high scalability andlow manufacturing cost, multi-accelerator systems widely exist in data centers,cloud platforms, and SoCs. Thus, a challenging problem arises inmulti-accelerator systems: selecting a proper combination of accelerators fromavailable designs and searching for efficient DNN mapping strategies. To thisend, we propose MARS, a novel mapping framework that can performcomputation-aware accelerator selection, and apply communication-aware shardingstrategies to maximize parallelism. Experimental results show that MARS canachieve 32.2% latency reduction on average for typical DNN workloads comparedto the baseline, and 59.4% latency reduction on heterogeneous models comparedto the corresponding state-of-the-art method.

MARS: Exploiting Multi-Level Parallelism for DNN Workloads on Adaptive Multi-Accelerator Systems