Initial Benchmarking of the Intel 3D-Stacked MCDRAM

Abstract

Modern, manycore processors are increasingly using high-bandwidth memory (HBM) to provide the necessary memory bandwidth to high core counts. The Knights Landing processor is one example, which is paired with a three-dimensional (3D) stacked memory, and a multi-channel direct random access memory (MCDRAM). The MCDRAM offers high-bandwidth memory, however, it is also important to understand the latency of the memory and how the memory will respond to noncontiguous accesses. This work provides information to programmers about how their applications will perform, and which of the many memory settings that will provide the best performance. These settings are non-trivial and can have a large impact on application performance. In addition to several micro-benchmarks, this work shows that using the MCDRAM can provide speedups up to 3.7x for a congregant gradient application, while giving slowdowns of 3x for certain hash table implementations. Given this range of performance, it is essential for programmers to understand this memory technology and how to use it appropriately.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Sep 20, 2019
Accession Number
AD1080853

Entities

People

  • Benjamin S. Parsons

Organizations

  • Engineer Research and Development Center

Tags

Communities of Interest

  • Materials and Manufacturing Processes

DTIC Thesaurus Topics

  • Army
  • Army Corps Of Engineers
  • Bandwidth
  • Case Studies
  • Central Processing Units
  • Classification
  • Computer Programming
  • Computers
  • Computing Devices
  • Computing System Architectures
  • Department Of Defense
  • Ecology
  • Engineering
  • Engineers
  • Hash Tables
  • Instructions
  • Measurement
  • Military Engineering
  • Performance Tests
  • Software Development
  • Specifications
  • Standards
  • Three Dimensional
  • Two Dimensional

Fields of Study

  • Computer science

Readers

  • Computer Engineering
  • Computer Networking
  • Mycotoxin ecology in Amazonian ecosystems.