Hi all,
I enabled an A100 GPU accelerator and used .set_memory to 120 gb (system + RAM on the GPU) and Google is confirming that my notebook is not using the extra RAM. It caps at 80, the system GPU, so I wanted to check that it wasn’t a problem on Psi4’s end detecting the wrong computational limit. I saw this linked thread on problems with parallelization but I don’t know if it’s related or fixed somehow:
Please note I’m terminating the calculation at the core.wavefunction() assembly, but that doesn’t seem relevant if it’s arbitrarily spiking much past what it would on a local runtime.