Test ridft

[wojek]

Hi all,

I've performed TM tests recently and all test passed but one. The one that fails (on 2 different computers) is located here: ./TURBOTEST/ridft/long/PB5.COSMO.GHF/ i'm getting this result:

PERFORMING COMMANDS:
/exports/local/TURBOMOLE602/bin/em64t-unknown-linux-gnu/ridft  > output.test 2> output.err

Total CPU time : 1386 s
Total wall time: 1387 s

TEST RESULTS:
CRITERION NAME                 TYPE RESULT                  REFRESULT           SUCCESS

Total energy                   c    -947.98321694        -966.69004640       failed
Number of iterations           c                      ---             44                  nogrep

TOTAL ENERGY NOT CORRECT.
NUMBER OF ITERATIONS NOT CORRECT.

THERE IS A PROBLEM WITH /exports/home/wojek/TURBOTEST/ridft/long/PB5.COSMO.GHF.

I'm using Turbomole 6.02. So ridft is already patched AFAIK. Do you probably know what is wrong?

best regards
Wojtek

[uwe]

Hello,

this problem occurs on certain Intel CPUs only. The combination of COSMO and spin-orbit coupling calculations seem to generate numerical problems with the used Intel Math Kernel library (MKL) that we use. Must be something with the linear algebra of complex matrices. We have linked Turbomole with an older version of MKL and that worked correctly also on the affected CPUs. So if you want to use COSMO with a spin-orbit calculation, please contact the Turbomole support, and you will get a special version of ridft. We recommend the usage of the default binaries in all other cases, however.

Users have reported those problems on Intel E54XX CPUs and on one Xeon 6000 with various CPU frequencies.

Please send an email to the Turbomole support with the output of

cat /proc/cpuinfo

of your machines to check if your CPU types match our findings.

Regards,

Uwe

[bangert]

Hi Uwe,

i can confirm the same problem on our new cluster.
here is our /proc/cpuinfo

processor       : 3
vendor_id       : GenuineIntel
cpu family      : 6
model           : 26
model name      : Intel(R) Xeon(R) CPU           L5520  @ 2.27GHz
stepping        : 5
cpu MHz         : 2265.800
cache size      : 8192 KB
fpu             : yes
fpu_exception   : yes
cpuid level     : 11
wp              : yes
flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss syscall nx rdtscp lm constant_tsc ida nonstop_tsc pni cx16 popcnt lahf_lm
bogomips        : 4533.26
clflush size    : 64
cache_alignment : 64
address sizes   : 40 bits physical, 48 bits virtual
power management: [8]

note: we run on centos5.3 on top of vmware esxi 4.
kind regards
Thilo

[uwe]

Hello,

just in case other users having that problem are reading this post here:

The problem has been solved, Intels newest MKL library fixed it and new binaries are available from the Turbomole support. Just send a short note to COSMOlogic to get the new version.

Regards,

Uwe