GPAI - General Purpose AI Computing

About

AI hardware accelerators are starting to be commonplace today and we foresee they will evolve by integrating more general- purpose friendly features, similar to the way GPUs evolved into GPGPUs more than a decade ago. Contemporary accelerators are highly parallel, specialized, and often limited-precision devices with higher performance per Watt than general CPUs and they are targeting a relatively narrow application domain, i.e., neural networks (deep learning) processing. We will address the challenges involved when using these AI accelerators for more general processing by (i) novel hardware and system software techniques, (ii) suitable computational models, and (iii) domain-specific programming models, languages and tools.

Funding

ELLIIT - Excellence Center at Linköping - Lund in Information Technology, project number C05, 2021-2025

Partners

People

BTH LiU LU
Prof. Håkan Grahn (PI)
Assist. Prof. Anton Borg
Christoffer Åleskog, Ph.D. student
Prof. Christoph Kessler (co-PI)
Dr. August Ernstsson
Sehrish Qummar, Ph.D. student
Assoc. Prof. Jörn Janneck (co-PI)
Assoc. Prof. Flavius Gruian
Michail Boulasikis, Ph.D. student
BTH Logo LiU Logo LU Logo