GPU QATK

[AsifShah]

Dear Admin,

I recently installed the X-2025.06 version of QuantumATK, which supports GPU acceleration for DFT calculations.
In the bin folder, I found the following files:
atkpython
atkpython_gpu
atkpython_system-mpi
quantumatk

In the mpi/bin, I found the following files:
cpuinfo
hydra_bstrap_proxy
hydra_nameserver
hydra_pmi_proxy
impi_info
mpiexec
mpiexec.hydra

Kindly let me know if this is how to run DFT simulations on GPU:

atkpython_gpu input.py > output.log

or

mpiexec.hydra -n 2 atkpython_gpu input.py > output.log (2 here for using 2 GPUs on a given node)
My GPU node shows 6 GPU machines when using nvidia-smi command.

If I am doing it incorrectly, kindly help me do it properly for single-GPU and multiple-GPU machines on a given node.

Also, do we require any licenses to use GPU acceleration because I tried atkpython_gpu input.py > output.log and it shows:
License Error: Unable to obtain a required license for the
requested license feature. It could mean there are no valid
licenses left, a problem with the connection to the license
server, or the license file does not contain the requested feature.
See below for more details.
No such feature exists.
Feature:       qatk_gpu
License path:  [email protected]:
FlexNet Licensing error:-5,234

[AsifShah]

Update: I tried re-installing new SCL license but the issue still remains. It gives the same error.

[Anders Blom]

Please contact me directly by email with your site ID so we can check.

[AsifShah]

Thanks, Anders Blom,

It's resolved now. The error mentioned that it was not recognizing "qatk_gpu" when running a GPU simulation. Later, I found that the older version of Smart Keys didn't have any description of "qatk_gpu"
So, had to update the smart keys, and it worked nicely.

I might have some questions regarding the speedup. I am still testing it.

For someone facing the same issues, kindly redownload your SmartKeys from the Solvnet and replace them with the old ones.

[Anders Blom]

Good! I thought you had done that already since you indicated "re-installing new SCL license". But yes, to be clear for all users, to use GPU you need to download an updated license from SolvNet/SmartKeys. All academic customers under active maintenance should have access to the GPU features after updating the license.