Login
Talks & Seminars
Title: Understanding the cost and the opportunity to enhance programmability of heterogeneous processors
Dr. Arkaprava Basu, AMD Research, Austin
Date & Time: June 27, 2017 10:30
Venue: Conference Room, C Block, 01st Floor, Dept. of CSE, Kanwal Rekhi (KReSIT) Bldg
Abstract:
As the benefits from transistor scaling diminish, the computing is turning increasingly heterogeneous. In a heterogeneous processor domain-specific accelerators like graphics processing unit (GPU) are coupled with general purpose CPUs for better performance and energy efficiency. However, programming such heterogeneous processors, consisting of multiple disparate computing element, is a challenge. In this talk, we will discuss two emerging features that can address this -- namely, (1) the shared virtual memory, and (2) the ability to invoke operating system services from an accelerator (e.g., GPU). Shared virtual memory or SVM makes it easy to share data across the CPU and accelerators by presenting them with the same view of the memory. Here, we will discuss possible performance pitfall of the SVM implementation in a heterogeneous processor and the opportunities to enhance its future iterations. Next, we will discuss how an accelerator's ability to invoke operating system services like page fault, file access, is crucial to make it a true peer to the CPU. We will then discuss the performance implications of this capability and avenues for betterment and further exploration. Finally, we will conclude the talk by discussing several future research opportunities to embrace heterogeneity across the hardware and the software stack and across the compute and the memory.
Speaker Profile:
Arkaprava (Arka) is a researcher with AMD Research at Austin,TX. Arka’s research interest broadly lies in the area of computer architecture. Since joining AMD Research in January of 2014, Arka has been exploring various aspects of virtual memory management techniques and heterogeneous processing as part of US Department of Energy’s Exascale computing initiative. Before that he obtained his PhD in Computer Science from the University of Wisconsin-Madison under the supervision of Prof. Mark D. Hill and Prof. Michael M. Swift in December, 2013. Prior to that Arka obtained M.Tech in computer science from IIT Kanpur and B.Tech from Government Engineering College in Kalyani, West Bengal.
List of Talks

Webmail

Username:
Password:
Faculty CSE IT
Forgot Password
    [+] Sitemap     Feedback