HPC Technology Survey 2021: Processing Technologies— CPUS and Accelerators

EXECUTIVE SUMMARY

EXECUTIVE SUMMARY

Intersect360 Research surveyed the High Performance Computing (HPC) user community on a wide range of technology issues. The complete study analyzes users’
current computing systems, storage systems, networks, cloud computing usage, and selected purchasing criteria. Our goal in this analysis is to provide
an overview of how HPC systems are configured, including the breadth of technologies most commonly used.

Intersect360 Research defines HPC as the use of servers, clusters, and supercomputers—plus associated software, tools, components, storage, and services—for
scientific, engineering, or analytical tasks that are particularly intensive in computation, memory usage, or data management. Intersect360 Research
reports available in this HPC Technology Survey report series include the following segmentations:

  • Server Technologies and Configurations: including system vendors installed, segmentation of systems into classes (entry-level, midrange, high-end,
    supercomputer), system interconnects, and usage of liquid cooling
  • Storage Technologies and Configurations: including total active HPC data; storage configurations spanning on-node, attached storage arrays,
    and cloud storage; parallel file system usage; and interdependence of storage and server elements for purchasing
  • Processing Elements—CPUs and Accelerators: including installations and user preferences for both CPUs and accelerators, independently
    and in combination, plus configuration options and interconnects
  • Cloud Computing: including proportion of computing and storage in public cloud for HPC and top named cloud vendors

This report provides a detailed examination of processing elements, including both CPUs and accelerators, installed among responding HPC user sites. This
analysis includes both a look at how processors are configured in HPC systems, and the reported vendors of both CPUs and accelerators. This includes
x86, ARM, and POWER processors plus Nvidia, AMD, and FPGA accelerators.

Intel CPUs still hold a commanding position in the market, with 98% of respondents reporting at least some usage and 88% reporting broad usage. AMD processors
are in use in 70% of HPC data centers, with nearly 25% relying on AMD CPUs broadly. For the first time in our surveys, ARM processor usage has surpassed
that of IBM’s POWER processor, with 24% using ARM versus 20% using POWER.

On the accelerator part of the survey, nearly 80% of the sites surveyed are using some type of accelerator, typically GPUs provided by Nvidia. AMD Instinct
GPUs were in use in 15% of the organizations surveyed while FPGAs were in use in just over 20% of respondent data centers. The number of accelerators
per node is probably larger than most believe, with 69% reporting four accelerators per node (slightly higher than the proportion using two accelerators
per node), and a surprising 38% of respondents using eight accelerators per node. Intersect360 Research attributes this trend to the expansion of machine
learning in HPC environments.

TABLE OF CONTENTS

EXECUTIVE SUMMARY 2

TABLE OF CONTENTS 4

INTRODUCTION 5

Methodology 5

What Is HPC? 6

Total Responses 6

Table 1: Total Estimated Clusters in Respondent Organizations 7

HPC PROCESSOR TECHNOLOGIES 7

Definitions 7

CPU Usage 8

Figure 1: CPU Usage in HPC Data Centers 8

HPC Accelerator Usage 9

Figure 2: Accelerator Type & Usage 9

Table 2: Accelerator Use by HPC Market Segment 11

Figure 3: Accelerators Per Node 11

Table 3: Accelerator Interfaces 12

CPU-Accelerator Combinations 12

Figure 4: CPU-Accelerator Combinations 13

CONCLUSIONS 14

APPENDIX A: SURVEY DEMOGRAPHICS 15

Table A1: Geographic Region of Respondent 15

Table A2: Economic Sector of Respondent Organization 15

Table A3: Number of Employees for Respondent Organization 15

Figure A1: Role of Survey Respondent in Organization 16

Figure A2: HPC Purchasing Responsibility of Respondent 16

Figure A3: Total Annual HPC Budget of Responding Organization 17

APPENDIX B: SELECTED QUALITATIVE COMMENTS ON ACCELERATORS 18