ERROR: GPU library not compiled for this accelerator

Hi users,

I am trying to use GPU capability of lammps on my laptop with Mac operating system. Specifically I tried GPU package installation following instructions available.

Following Makefile.mac with
CUDA_PRECISION = -D_DOUBLE_DOUBLE (As having Nvidia GT750M with CC3.0)

CUDA_ARCH = -arch=sm_35 (Tried 21,30 too)

NVCC = nvcc -m64

Able to get libgpu.a successfully for all three sm_21/30/35 values (which I explored from lammps mailing list posts for similar posts)

Still the error looks persistent. I will appreciate if you guys can give some suggestion to get this issue resolved. I am pasting the outcome of ./nvc_get_devices below

For the GT750M card, -arch=sm_30 is the correct compute capability setting. After modifying Makefile.mac, did you do “make -f Makefile.mac clean” and then “make -f Makefile.mac”?

Once libgpu.a is rebuilt, the LAMMPS binary should be re-linked with updated libgpu.a: i.e. in the src/ folder

rm lmp_foo; make foo

Hope it helps,
-Trung

Hi Trung,

Thanks for your advice.
I did it all from scratch for arch_sm=30 and now it is running benchmark and gpu example codes.
Though I did it previously but seems I forgot something in between.

A small query just for curiosity. What helps in deciding “arch_sm” value based on GPU card provided ?

Thanks for resolving the problem.

A small but good answer is goven on SO:
http://stackoverflow.com/questions/10330527/what-is-difference-between-arch-sm-13-and-arch-sm-20/10331886#10331886

Regards,

M.