That error message is self-explanatory. Check out the documentation for the package command.
Try running inputs from the bench folder, and use the variable settings for x y and z to increase the system size. There is very little acceleration to be had for small systems.
Here is a workaround for avoiding the compilation failure with pathnames containing blanks.
No, I am not putting space ever again while naming folder. I would rather put underscore ( _ )
It uses only 8% gpu. I don’t know if this is okay. Here is the screen recording you might want to have a look -
https://jumpshare.com/share/jTI9GwwzIlutGTmlAEcc
I was running in.spce.hex file in HEAT example
No surprise there. This is a tiny system.
1 Like
But I agree with @initialize, the fact that the hardware does not have any double precision support is a serious restriction. Since the KOKKOS support for single precision is very new, having a single precision-only hardware could have some challenges. I recall that running an Intel Xe generation GPU (embedded in an Intel CPU on a laptop) took some effort to detect and remove a few unintentional uses of double precision math (which leads to the GPU kernels crashing).
Yes the KOKKOS package still uses some double precision (FP64), even when compiling with SINGLE_SINGLE. Your GPU really can’t use emulation for FP64 (even if it is slow)??
1 Like
So GPU can crank up to 100% if needed?
Thank you Mr. Kohlmeyer. I really appreciate you for helping me. You were really apt to get me help. It was nice having conversation with you 
So double precision is a hardware oriented and it doesn’t come up with software. So I am not getting any support in future. Correct me If I am wrong?
