I noticed that there was higher GPU usage (78%) for the CXL_DIM than the other Gerald Wu's (67%) for a windows 7 computer, when the WU is running on a 980 Ti card.

https://www.gpugrid.net/result.php?resultid=14608791


See link below.

Please note: In boinc manager device 0 is the 980 Ti card, while device 1 and 2 is the 690 card. In afterburner GPU 1 is the 980 Ti, while GPU 2 & 3 is the 690 card. The work unit was running on the 690 card then I reboot the computer and it finished on the 980 Ti, then I noticed this.

I hope the links below work.


https://www.dropbox.com/s/vhh1ncg4vplhzdu/CXL12_DIM%20a1.jpg?dl=0





https://www.dropbox.com/s/xund294zskoawdb/VACXCL12%20b1.jpg?dl=0


Is why is this happening? What is different about this WU type compared to the others?

My windows xp computer was able to achieve 97% GPU usage with 980 Ti card, which is slightly better than the 93%, I usually get.

See link ( I hope it opens):

https://www.dropbox.com/s/ivzp65o076y3mxa/CXCL12_DIM_HEP1%20b1.jpg?dl=0



See unit link:


https://www.gpugrid.net/result.php?resultid=14612204


These WUs (CXL12_DIM) definitely have a higher GPU usage.

I noticed that there was higher GPU usage (78%) for the CXL_DIM than the other Gerald Wu's (67%) for a windows 7 computer, when the WU is running on a 980 Ti card.
...
Is why is this happening? What is different about this WU type compared to the others?

The difference is in the number of atoms in the given simulation.
Look for the number before "Natoms" at the last line of stderr.txt:

GERARD_CXCL12_DIM_HEP1    : 64965 Natoms 10.000.000 steps
GERARD_CXCL12_CMP_11_mol1 : 29570 Natoms 15.000.000 steps
GERARD_CXCL12_CMP_02_mol2 : 29572 Natoms 15.000.000 steps
GERARD_PTCL2_CTL_PRZ3     : 37886 Natoms 15.000.000 steps
GERARD_PTCTL_PLA2_AIN3    : 29531 Natoms 15.000.000 steps
GERARD_PTCTL_PLA2_IBP3    : 29522 Natoms 15.000.000 steps
GERARD_PTCTL_LFE_IBP1     : 54751 Natoms 11.250.000 steps
GERARD_FXCXCL12_LIG       : 31843 Natoms 17.500.000 steps
GERARD_VACXCL12_LIG       : 32030 Natoms 17.500.000 steps
GERARD_3GG_CXCL12_DIM     : 35159 Natoms  5.000.000 steps (short)

The larger the number of atoms, the longer one step takes, so less interaction is needed between the CPU and the GPU to keep the simulation running. Less interaction between the CPU and the GPU results in less latency caused by the WDDM, thus the GPU usage will be higher.
Workunits with more atoms consist less steps, to keep the total runtime similar.
However the ns/day value is wrong, it is either day/ns or ms/step.