GPU acceleration on RTX 2060 ?? windows version


On older and the latest win version of lammps gpu accel dont works with new RTX 2060 card or drivers.
Before it was GTX 770 and no problems.

LAMMPS (19 Sep 2019)
OMP_NUM_THREADS environment is not set. Defaulting to 1 thread. (…/comm.cpp:93)
using 1 OpenMP thread(s) per MPI task
ERROR: Could not find/initialize a specified accelerator device (…/gpu_extra.h:35)
Last command: package gpu 1

Found 0 platform(s).

I tried clear driver installation with DDU and safe mode.
I tried replacing nvopencl.dll as here advised

Latest nvidia cuda driver

OpenCL geekbench works fine.

There have been similar reports of this already and there is - unfortunately - no recommendation or suggestion to resolve this. There have also been reports of problems with the newer OpenCL 2.x drivers from AMD. However, outside of acknowledging the situation, there is little that can be done about it, currently.

The windows binaries are built using the reference ICD dynamic back-end loaded API as it was available with the OpenCL 1.2 sources.
The code has not been changed since, so there must be a change in OpenCL interface of the CUDA drivers (like with the AMD drivers).
Unfortunately, I have no way (and no time) to debug and resolve this. The LAMMPS windows binaries are built with a cross-compiler on Linux. I have only one desktop available with a working windows license and that has no nvidia GPU (and neither can one be added). I am using this machine in windows mode very rarely (like once ever 3-4 months). but without nvidia hardware (or AMD hardware for that matter) it is impossible to test and debug what is wrong there.

We (the LAMMPS developers) would be more than happy to hand over the support for the windows platform to somebody with better suited equipment and more time and experience on windows and would be willing to also collaborate to work on a port to the native windows compilers from microsoft to also add support for cuda and have properly signed executables, which would allow people to use the binaries on newer windows machines without having to have administrator privileges.

Until then, we can only recommend to turn your machine into a dual-boot box with a linux partition, where GPU and CUDA support is easier to achieve and better.