Order of magnitude timing difference

Message

tjf · #1 Post by **tjf** » Fri Sep 30, 2005 2:12 pm

Does anyone have a decent explanation for the folliwing change in the timing of two near-identical VASP runs:

The two runs differ in one lattice constant (by around 2%) and small changes (<1%) in the atomic coordinates (same symmetry). Timings for the first iteration:

Run 1 LOOP: VPU time 1082.79: CPU time 1083.68
Run 2 LOOP: VPU time12503.74: CPU time12520.91

The biggest difference comes from EDDAV, which goes from 1079.02 to 12497.18. That's an order of magnitude change in the timing!!!

Diffing the OUTCARs shows the only changes are minor changes (~2%) in IRMAX, IRDMAX, NGXF, NGX, # planewaves and k-point Cartesians due to change in lattice parameter.

Ideas?

#2 Post by **admin** » Wed Oct 12, 2005 11:40 am

please check how many minimisation steps were needed
in both steps (grep eigenvalue-minimisations in OUTCAR)

tjf · #3 Post by **tjf** » Mon Oct 17, 2005 3:25 pm

please check how many minimisation steps were needed

No significant difference.

I take it from the lack of replies that there's no obvious answer?

Here's output for the first iteration:

Run 1:

----------------------------------------- Iteration 1( 1) ---------------------------------------

POTLOK: VPU time 2.90: CPU time 2.90
SETDIJ: VPU time 0.09: CPU time 0.09
EDDAV : VPU time 1079.02: CPU time 1079.86
BZINTS: Fermi energy: 8.205941; 80.000000 electrons
Band energy:-591.438413; BLOECHL correction: -0.084118
DOS : VPU time 0.00: CPU time -0.03
------------------------------------------
LOOP: VPU time 1082.79: CPU time 1083.68

eigenvalue-minimisations : 15968
total energy-change (2. order) : 0.5059596E+03 (-0.4840716E+04)
number of electron 80.0000000 magnetization
augmentation part 80.0000000 magnetization

Free energy of the ion-electron system (eV)
---------------------------------------------------
alpha Z PSCENC = 89.66104942
Ewald energy TEWEN = -1057.77638328
-1/2 Hartree DENC = -567.62293893
-V(xc)+E(xc) XCENC = 232.94636182
PAW double counting = 587.79977387 -784.30323160
entropy T*S EENTRO = 0.00000000
eigenvalues EBANDS = -591.43841334
atomic energy EATOM = 2596.69338122
---------------------------------------------------
free energy TOTEN = 505.95959919 eV

energy without entropy = 505.95959919 energy(sigma->0) = 505.95959919

Run 2:
----------------------------------------- Iteration 1( 1) ---------------------------------------

POTLOK: VPU time 5.72: CPU time 5.72
SETDIJ: VPU time 0.13: CPU time 0.13
EDDAV : VPU time12497.18: CPU time12514.31
BZINTS: Fermi energy: 8.058645; 80.000000 electrons
Band energy:-603.535796; BLOECHL correction: -0.062359
DOS : VPU time 0.00: CPU time -0.03
------------------------------------------
LOOP: VPU time12503.74: CPU time12520.91

eigenvalue-minimisations : 16000
total energy-change (2. order) : 0.5003464E+03 (-0.4830743E+04)
number of electron 80.0000000 magnetization
augmentation part 80.0000000 magnetization

Free energy of the ion-electron system (eV)
---------------------------------------------------
alpha Z PSCENC = 87.95321994
Ewald energy TEWEN = -1035.19320215
-1/2 Hartree DENC = -581.74469477
-V(xc)+E(xc) XCENC = 232.67690316
PAW double counting = 587.79977387 -784.30323160
entropy T*S EENTRO = 0.00000000
eigenvalues EBANDS = -603.53579622
atomic energy EATOM = 2596.69338122
---------------------------------------------------
free energy TOTEN = 500.34635345 eV

energy without entropy = 500.34635345 energy(sigma->0) = 500.34635345