Fermilab Computing Division

CS Document 5406-v3

CMS Intel Phi Investigation

Document #:
Document type:
Technical Note
Submitted by:
Stephan Lammel
Updated by:
Stephan Lammel
Document Created:
19 Aug 2014, 11:53
Contents Revised:
25 Aug 2014, 12:42
Metadata Revised:
25 Aug 2014, 12:42
Viewable by:
  • Public document
Modifiable by:

Quick Links:
Latest Version

Other Versions:
19 Aug 2014, 16:31
19 Aug 2014, 11:57
19 Aug 2014, 11:55
Intel's current Xeon Phi coprocessor cards provide up to a
thousand GFLOPS of double precision performance, i.e. about
four times current server processor performance. Investigating
the use of the technology for compute intensive high-energy-
physics applications is thus natural.

The Compact Muon Solenoid experiment observes proton--proton
collisions at a rate of $40 \, \mbox{MHz}$, records about $2 \,
\mbox{PetaBytes}$ of data per year, and utilizes around $100,000$
CPU cores at any given time to reconstruct, simulate, and analyze
this data.

A first investigation of the technology run into obstacles
compiling the complex reconstruction, simulation, and analysis
software of the experiment with the Intel C++ compiler. The
experiment software uses extensions provided by the C++11 standard
that the Intel compiler does not (yet) support. A stand-alone
Monte Carlo generator, capable of utilizing many cores, showed
an Intel Xeon Phi core to perform significantly slower than a
current server processor core.

Files in Document:
DocDB Home ]  [ Search ] [ Authors ] [ Events ] [ Topics ]

DocDB Version 8.8.9, contact Document Database Administrators