Difference between revisions of "Benchmarking: HEPSPEC"

From Define Wiki
Jump to navigation Jump to search
 
(16 intermediate revisions by 4 users not shown)
Line 10: Line 10:
 
# CentOS 6
 
# CentOS 6
 
yum groupinstall 'Development tools'
 
yum groupinstall 'Development tools'
yum install glibc-devel-2.12-1.47.el6.i686 compat-gcc-34-g77
+
#OLD:yum -y install glibc-devel-2.12-1.47.el6.i686 compat-gcc-34-g77
 +
yum -y install glibc-devel.i686 compat-gcc-34-g77
 +
yum -y install libstdc++-devel.i686 sharutils bc mailx
  
 
# CentOS 5
 
# CentOS 5
yum groupinstall 'Development tools'
+
yum -y groupinstall 'Development tools'
yum install glibc-devel compat-gcc-34-g77
+
yum -y install glibc-devel compat-gcc-34-g77
  
 
## JH Test
 
## JH Test
yum install glibc-devel.i686 libstdc++.i686 sharutils gcc-c++
+
yum -y install glibc-devel.i686 libstdc++.i686 sharutils gcc-c++
yum install compat-libstdc++-33.i686 libstdc++-devel.i686 compat-libcap1.i686  
+
yum -y install compat-libstdc++-33.i686 libstdc++-devel.i686 compat-libcap1.i686  
 
</syntaxhighlight>
 
</syntaxhighlight>
  
Line 55: Line 57:
 
# as root
 
# as root
 
./runspec.sh -d "Benchmark HEPSPEC2006 32bit" -a 32 -b all_cpp
 
./runspec.sh -d "Benchmark HEPSPEC2006 32bit" -a 32 -b all_cpp
 +
 +
# form hepix site:
 +
./runspec.sh -d "HEP-SPEC06 32-bit" -a 32 -b all_cpp -e your.email@address.com
 
</syntaxhighlight>
 
</syntaxhighlight>
  
Line 70: Line 75:
  
 
=== Intel Results ===
 
=== Intel Results ===
 +
http://wiki.bostonlabs.co.uk/w/index.php/Results:Hepspec
 +
 +
=== AMD Results ===
  
 
{| class="wikitable" style="text-align:center; width:60%; "
 
{| class="wikitable" style="text-align:center; width:60%; "
 
|+ HEPSPEC2006 results for Dual CPU system
 
|+ HEPSPEC2006 results for Dual CPU system
 
|-
 
|-
! scope="col" | CPU !! scope="col" | Freq !! scope="col" | Cores !! scope="Cores" | Hepspec (no HT) !! scope="col" | Hepspec (with HT)
+
! scope="col" | CPU !! scope="col" | Freq !! scope="col" | CPU !! scope="Cores" | Hepspec
 
|-
 
|-
| colspan="5"|''Sandy Bridge''
+
| colspan="4"|''Rome (Data provided by Marlon Cunningham from AMD (05/03/2020))''
 
|-
 
|-
! scope="row" | E5-2687W || 3.10GHz || 8 || 346.97 || 419
+
! scope="row" | Epyc 7252 || 3.10GHz || 32 (SMT on) || 523
 
|-
 
|-
! scope="row" | E5-2690 || 2.90GHz || 8 || 324.58 || 392
+
! scope="row" | Epyc 7262 || 3.20GHz || 32 (SMT on) || 589
 
|-
 
|-
! scope="row" | E5-2680 || 2.70GHz || 8 || 291.13 || 352
+
! scope="row" | Epyc 7272 || 2.90GHz || 48 (SMT on) || 729
 
|-
 
|-
! scope="row" | E5-2670 || 2.60GHz || 8 || 246.20 || 346
+
! scope="row" | Epyc 7282 || 2.80GHz || 64 (SMT on) || 909
 
|-
 
|-
! scope="row" | E5-2665 || 2.40GHz || 8 || 268.62 || 320.86
+
! scope="row" | Epyc 7302 || 3.00GHz || 64 (SMT on) || 1029
 
|-
 
|-
! scope="row" | E5-2660 || 2.20GHz || 8 || 246.24 || 297
+
! scope="row" | Epyc 7352 || 2.30GHz || 96 (SMT on) || 1279
 
|-
 
|-
! scope="row" | E5-2650 || 2.00GHz || 8 || 223.85 || 270
+
! scope="row" | Epyc 7402 || 2.80GHz || 96 (SMT on) || 1427
 
|-
 
|-
! scope="row" | E5-2650L || 1.80GHz || 8 || || 239.17
+
! scope="row" | Epyc 7502 || 2.50GHz || 128 (SMT on) || 1682
 
|-
 
|-
! scope="row" | E5-2640 || 2.50GHz || 6 || 196.39 || 237
+
! scope="row" | Epyc 7542 || 2.90GHz || 128 (SMT on) || 1791
 
|-
 
|-
! scope="row" | E5-2630 || 2.30GHz || 6 || 180.68 || 218
+
! scope="row" | Epyc 7552 || 2.20GHz || 196 (SMT on) || 2270
 
|-
 
|-
! scope="row" | E5-2620 || 2.00GHz || 6 || 157.11 || 190
+
! scope="row" | Epyc 7642 || 2.30GHz || 196 (SMT on) || 2359
 
|-
 
|-
| colspan="5"|''Westmere/Nehalem''
+
! scope="row" | Epyc 7702 || 2.00GHz || 256 (SMT on) || 2562
 
|-
 
|-
! scope="row" | X5680 || 3.33GHz || 6 || 204.03 ||  
+
! scope="row" | Epyc 7742 || 2.25GHz || 256 (SMT on) || 2759
 
|-
 
|-
! scope="row" | X5670 || 2.93GHz || 6 || 181.15 ||
+
! scope="row" | Epyc 7H42 || 2.6GHz || 256 (SMT on) || 3053
|-
 
! scope="row" | X5660 || 2.80GHz || 6 || 173.2 ||
 
|-
 
! scope="row" | X5650 || 2.66GHz || 6 || 164.46 ||
 
|-
 
! scope="row" | E5640 || 2.66GHz || 4 || 102.35 ||
 
|-
 
! scope="row" | E5630 || 2.53GHz || 4 || 97.35 ||
 
|-
 
! scope="row" | E5620 || 2.40GHz || 4 || 92.35 ||
 
|-
 
| colspan="5"|''Ivy Bridge''
 
|-
 
! scope="row" | E5-2650 V2 || || 10 || || 366.8
 
|-
 
! scope="row" | E5-2660 V2 || || 10 || || 405.4
 
|-
 
! scope="row" | E5-2690 V2 || || 10 || || 489.05
 
|-
 
|colspan="5"|''Haswell''
 
|-
 
! scope="row" | E5-2695 V3 || 2.30GHz || 14 || 365.66 || 531.85
 
|}
 
 
 
=== AMD Results ===
 
 
 
{| class="wikitable" style="text-align:center; width:60%; "
 
|+ HEPSPEC2006 results for Dual CPU system
 
|-
 
! scope="col" | CPU !! scope="col" | Freq !! scope="col" | CPU !! scope="Cores" | Hepspec
 
 
|-
 
|-
 
| colspan="4"|''Abu Dhabi''
 
| colspan="4"|''Abu Dhabi''

Latest revision as of 15:59, 5 February 2020

Copy over the files

  • Note: Use SPEC2006_v12.tar.bz2

Copy over the spec2k6 files from the shared directory: //10.0.0.220/software/hpc/benchmarks/SPEC2006_v12.tar.bz2

NOTE Place all files in the same directory

32 bit libs

Make sure you have the 32bit libs installed!

# CentOS 6
yum groupinstall 'Development tools'
#OLD:yum -y install glibc-devel-2.12-1.47.el6.i686 compat-gcc-34-g77
yum -y install glibc-devel.i686 compat-gcc-34-g77
yum -y install libstdc++-devel.i686 sharutils bc mailx

# CentOS 5
yum -y groupinstall 'Development tools'
yum -y install glibc-devel compat-gcc-34-g77

## JH Test
yum -y install glibc-devel.i686 libstdc++.i686 sharutils gcc-c++
yum -y install compat-libstdc++-33.i686 libstdc++-devel.i686 compat-libcap1.i686

Download the CERN HEPSPEC configuration file

# check https://twiki.cern.ch/twiki/bin/view/FIOgroup/TsiBenchHEPSPEC for the latest version
# OLD: wget --no-check-certificate https://twiki.cern.ch/twiki/pub/FIOgroup/TsiBenchHEPSPEC/spec2k6-2.14.tar.gz
wget --no-check-certificate https://twiki.cern.ch/twiki/pub/FIOgroup/TsiBenchHEPSPEC/spec2k6-2.23.tar.gz

Extract the spec config files

NB: When running HEPSPEC for a CERN configuration with the default CERN filesystem layout, do not attempt to run HEPSPEC from within "/" or "/root/". The default partition size is not big enough, the partition will run out of space and the benchmark will fail. Use "/pool/" instead.

david@cal1:~/$ mkdir -p scratch/hepspec && cd scratch/hepspec
david@cal1:~/scratch/hepspec$ mv ~/spec2k6-2.14.tar.gz .
david@cal1:~/scratch/hepspec$ tar zxvf spec2k6-2.14.tar.gz  
linux32-gcc_cern.cfg
linux64-gcc_cern.cfg
README.TXT
runspec.sh
simple.sh

Edit the runspec.sh file

NOTE 1 Make sure this is done or results get sent to cern (if email is configured on the server)
NOTE 2 The RESULTDIR variable must point to a directory that exists (the script won't create it). If it doesn't exist, the results won't be saved!

# vi runspec.sh
EMAIL="hpc@boston.co.uk"
RESULTDIR="/home/david/scratch/hepspec/results-spec"

Run HEPSPEC

# as root
./runspec.sh -d "Benchmark HEPSPEC2006 32bit" -a 32 -b all_cpp

# form hepix site:
./runspec.sh -d "HEP-SPEC06 32-bit" -a 32 -b all_cpp -e your.email@address.com

Note for multiple runs

NOTE Extracting the tar.bz2 files takes some time. If the build fails after the archive is extracted, remove the following line from runspec (around line 75) and rerun the build

tar xjf $SPECTARBALL || fail "Unable to extract SPECint!"

And remove or comment out the line near the end of the script (around line 190) which deletes the extracted spec files

rm -rf $SPECDIR

Results

Intel Results

http://wiki.bostonlabs.co.uk/w/index.php/Results:Hepspec

AMD Results

HEPSPEC2006 results for Dual CPU system
CPU Freq CPU Hepspec
Rome (Data provided by Marlon Cunningham from AMD (05/03/2020))
Epyc 7252 3.10GHz 32 (SMT on) 523
Epyc 7262 3.20GHz 32 (SMT on) 589
Epyc 7272 2.90GHz 48 (SMT on) 729
Epyc 7282 2.80GHz 64 (SMT on) 909
Epyc 7302 3.00GHz 64 (SMT on) 1029
Epyc 7352 2.30GHz 96 (SMT on) 1279
Epyc 7402 2.80GHz 96 (SMT on) 1427
Epyc 7502 2.50GHz 128 (SMT on) 1682
Epyc 7542 2.90GHz 128 (SMT on) 1791
Epyc 7552 2.20GHz 196 (SMT on) 2270
Epyc 7642 2.30GHz 196 (SMT on) 2359
Epyc 7702 2.00GHz 256 (SMT on) 2562
Epyc 7742 2.25GHz 256 (SMT on) 2759
Epyc 7H42 2.6GHz 256 (SMT on) 3053
Abu Dhabi
6380 x.xxGHz xx 286.4
Interlagos
6282 x.xxGHz xx 308.82
6276 x.xxGHz xx 288.90
6274 x.xxGHz xx 268.98
6272 x.xxGHz xx 259.01
6238 x.xxGHz xx 179.32
6234 x.xxGHz xx 171.85
6220 x.xxGHz xx 109.58
6212 x.xxGHz xx 99.62
Magney-Cours
6174 2.20GHz 12 195.42
6172 2.10GHz 12 186.54
6168 1.90GHz 12 168.77
6136 2.40GHz 8 142.12
6134 2.30GHz 8 136.2
6128 2.00GHz 8 118.44