DATE 2019 Special Day: Embedded Meets Hyperscale and HPC

In March 2019, Christian Plessl and Christoph Hagleitner co-organized a special day on the topic  "Embedded Meets Hyperscale and HPC" at the DATE 2019 conference. The objective of this day was to bring together leading experts and researchers from embedded, datacenter and high-performance computing to discuss what these fields can learn from each other. The event was very successful and attracted more than 100 participants. This website serves as the archive of the program and and talks of the speakers, who have agreed to make their slides available.

Summary

Heterogeneous computing with multiple, specialized processors and application-specific accelerators is vital for embedded systems to meet performance, latency, and efficiency targets. The same goals of fast, efficient, and cost-effective processing are also gating factors for the evolution of hyperscale data center (DC) and high-performance computing  (HPC) and Moore’s law no longer provides the necessary efficiency gains. Hence, we can witness a spread of technologies pioneered in embedded systems to hyperscale DCs and HPC systems, e.g., the use of specialized computing resources, massive parallelism, model-driven and domain-specific programming, task models, co-scheduling, and others. The theme of this special day is to highlight this confluence of methods and technologies to better understand, how heterogeneous computing is shaping the future of hyperscale DCs and HPC.

Program

Session: Heterogeneous Computing in the Datacenter and HPC

Talks

  • Silicon Heterogeneity in the Cloud
    Babak Falsafi, EPFL, Switzerland (AbstractSlides)
  • GPU Accelerated Computing in HPC and in the Datacenter
    Peter Messmer, NVidia, Switzerland (Abstract | Slides)
  • Heterogeneous Compute Architecture for Deep Learning in the Cloud
    Nicloas Fraser, Xilinx Research, Ireland (Abstract | Slides)

Session: Near Memory Computing

Talks

  • NTX: An Energy-Efficient Streaming Acclerator for Floating-Point Generalized Reduction Worksloads in 22nm FD-SOI
    Luca Benini, ETH Zurich, Switzerland and University of Bologna, Italy (Abstract | Slides)
  • Near-Memory Processing: It's the Hardware AND Software, Stupid!
    Boris Grot, University of Edinburgh, UK (Abstract | Slides)
  • Coherently Attached Programmable Near-Memory Acceleration Platform and its Application to Stencil Processing
    Jan van Lunteren, IBM Research, Switzerland (Abstract | Slides)

Keynote: Heterogeneous, High-Scale Computing in the Era of Intelligent, Cloud-Connected Devices

  • David Pellerin
    Amazon Webservices, US (Abstract | Slides)

Session: Tools and Runtime Systems

Talks

  • Extreme Heterogeneity in High Performance Computing
    Jeffrey Vetter, Oak Ridge National Laboratory, US (Abstract | Slides)
  • Homogenizing Heterogeneity: The OmpSs Approach
    Jesus Labarta, Barcelona Supercomputing Center, Spain (Abstract | Slides)
  • Automatic Code Restructuring for FPGAs: Current Status, Trends and Open Issues
    João M. P. Cardoso, University of Porto, Portugal (Abstract | Slides)

Panel Discussion: What Can HPC and Hyperscale Learn from Embedded Computing – and vice versa

Contact

Dr. Christoph Hagleitner
Manager Accelerator Technologies
IBM Research Zurich, Switzerland