Mozart: Taming Taxes and Composing Accelerators with Shared-Memory
Resource-constrained system-on-chips (SoCs) are increasingly heterogeneous with specialized accelerators for various tasks. Acceleration taxes due to control and data movement, however, diminish end-to-end speedups from hardware acceleration. Meanwhile, emerging workloads are increasingly task-diverse with several, potentially shared, fine-grained acceleration candidates. This motivates a paradigm of parallel and disaggregated acceleration. Compared to a monolithic accelerator, disaggregation provides higher flexibility, reuse, and utilization, but at the cost of higher control and data acceleration taxes.
Authors
- Zeran Zhu
- Luca P. Carloni (ORCID: https://orcid.org/0000-0001-5600-8931)
- Davide Giri (ORCID: https://orcid.org/0000-0003-4101-4516)
- Ying Jing (ORCID: https://orcid.org/0000-0001-5299-4746)
- Sarita V. Adve (ORCID: https://orcid.org/0000-0002-3403-5119)
- Paolo Mantovani (ORCID: https://orcid.org/0000-0002-1901-8732)
- Naiyin Jin (ORCID: https://orcid.org/0009-0009-5552-469X)
- Charles Block (ORCID: https://orcid.org/0009-0003-7770-003X)
- Vignesh Suresh (ORCID: https://orcid.org/0000-0003-4828-2759)
- Bakshree Mishra (ORCID: https://orcid.org/0000-0003-4904-0989)
- J D Zuckerman (ORCID: https://orcid.org/0000-0003-3081-1077)
Institutions
- University of Illinois Urbana-Champaign (US)
- Columbia University (US)
Publication Details
- Published
- 2024-10-11
- DOI
- https://doi.org/10.1145/3656019.3676896
- Citations
- 2
- Primary Topic
- Parallel Computing and Optimization Techniques
- Type
- article
- Field-Weighted Citation Impact
- 0.98