[Starpu-devel] StarPU cannot find GPUs on Plafrim

[Yihong Gao <yihong.gao@inria.fr>]

Hello, everybody,

  I'm running an experiment on Chameleon+StarPU. I use the Mirage server on Plafrim, but the StarPU cannot find the GPUs of the server. The results are as follows:

[ygao@mirage06 ~]$ $(spack location -i chameleon~simu)/lib/chameleon/timing/time_dgeqrf_tile --gpus=3 --trace --nb=960 --ib=96 --n_range=15360:15360:1 --niter=10
[starpu][starpu_initialize] Warning: StarPU was configured with --with-fxt, which slows down a bit, limits scalability and makes worker initialization sequential
[starpu][_starpu_init_machine_config] Warning: 3 CUDA devices requested. Only 0 available.
#
# CHAMELEON 0.9.1, /home/ygao/spacknew/opt/spack/linux-x86_64/gcc-4.8.4/chameleon-0.9.1-robmkyily4y2rrju5sndpfvptdjzv3gv/lib/chameleon/timing/time_dgeqrf_tile
# Nb threads: 9
# Nb GPUs: 3
# NB: 960
# IB: 96
# eps: 1.110223e-16
#
# M N K/NRHS seconds Gflop/s Deviation
15360 15360 1

  The configuration I used on the server is here:

module load slurm

module load compiler/gcc/4.8.4

module load build/cmake/3.2.1

module load hardware/hwloc/1.11.8

module load trace/fxt/0.3.3

module load intel/mkl/64/2017/update4

export LIBRARY_PATH=/usr/lib64:$LIBRARY_PATH

module load cuda80/toolkit/8.0.61

module load build/ac269-am114-lt246-m41417

export CPATH=/cm/shared/dev/apps/trace/fxt/0.3.3/include/:SCPATH 

 And the installation command is here:

spack install -v chameleon@0.9.1+cuda+starpu+fxt~mpi~simu ^starpu@trunk+cuda+fxt~mpi ^mkl ^cmake@exist ^cuda@exist ^hwloc@1.11.8

  I also use cuda75, the same problem still exists. How can I find my GPUs?

  Many thanks for your advice.

Best Regards,

Yihong Gao