Hi, Don't use those old version of libgpuarray and Theano. Use libgpuarray 0.6.2 and Theano 0.9rc3.
For the small GPU utilisation. A few candidate, you spend much time in python between your Theano function call and Theano isn't the bottleneck. If that isn't the case, maybe one op isn't on the GPU. You can profile Theano: http://deeplearning.net/software/theano/tutorial/profiling.html On Sat, Mar 4, 2017 at 4:10 AM Ragav Venkatesan <[email protected]> wrote: > When I run the make command, here is what I get: > > Scanning dependencies of target gpuarray > Scanning dependencies of target gpuarray-static > [ 1%] Building C object src/CMakeFiles/gpuarray-static.dir/cache/lru.c.o > [ 3%] Building C object > src/CMakeFiles/gpuarray-static.dir/gpuarray_error.c.o > [ 8%] Building C object > src/CMakeFiles/gpuarray-static.dir/gpuarray_array_blas.c.o > [ 5%] Building C object src/CMakeFiles/gpuarray-static.dir/cache/twoq.c.o > [ 6%] Building C object src/CMakeFiles/gpuarray.dir/gpuarray_util.c.o > [ 11%] Building C object src/CMakeFiles/gpuarray.dir/gpuarray_extension.c.o > [ 11%] Building C object > src/CMakeFiles/gpuarray.dir/gpuarray_buffer_opencl.c.o > [ 15%] Building C object src/CMakeFiles/gpuarray.dir/cache/lru.c.o > [ 16%] Building C object src/CMakeFiles/gpuarray.dir/gpuarray_strl.c.o > [ 18%] Building C object > src/CMakeFiles/gpuarray-static.dir/gpuarray_types.c.o > [ 20%] Building C object > src/CMakeFiles/gpuarray-static.dir/gpuarray_buffer.c.o > [ 13%] Building C object > src/CMakeFiles/gpuarray-static.dir/gpuarray_array.c.o > [ 22%] Building C object src/CMakeFiles/gpuarray.dir/gpuarray_types.c.o > [ 23%] Building C object > src/CMakeFiles/gpuarray.dir/gpuarray_array_blas.c.o > [ 25%] Building C object src/CMakeFiles/gpuarray-static.dir/util/strb.c.o > [ 27%] Building C object src/CMakeFiles/gpuarray.dir/gpuarray_kernel.c.o > [ 28%] Building C object > src/CMakeFiles/gpuarray-static.dir/gpuarray_util.c.o > [ 30%] Building C object > src/CMakeFiles/gpuarray-static.dir/gpuarray_blas_cuda_cublas.c.o > [ 32%] Building C object src/CMakeFiles/gpuarray.dir/gpuarray_buffer.c.o > [ 33%] Building C object > src/CMakeFiles/gpuarray-static.dir/gpuarray_extension.c.o > [ 35%] Building C object src/CMakeFiles/gpuarray.dir/gpuarray_error.c.o > [ 37%] Building C object > src/CMakeFiles/gpuarray-static.dir/gpuarray_elemwise.c.o > [ 38%] Building C object > src/CMakeFiles/gpuarray.dir/gpuarray_buffer_cuda.c.o > [ 40%] Building C object src/CMakeFiles/gpuarray.dir/cache/twoq.c.o > [ 42%] Building C object src/CMakeFiles/gpuarray.dir/gpuarray_elemwise.c.o > [ 44%] Building C object > src/CMakeFiles/gpuarray-static.dir/gpuarray_kernel.c.o > [ 45%] Building C object > src/CMakeFiles/gpuarray-static.dir/gpuarray_buffer_opencl.c.o > [ 47%] Building C object src/CMakeFiles/gpuarray.dir/gpuarray_array.c.o > [ 49%] Building C object > src/CMakeFiles/gpuarray-static.dir/gpuarray_strl.c.o > [ 50%] Building C object > src/CMakeFiles/gpuarray.dir/gpuarray_blas_cuda_cublas.c.o > [ 52%] Building C object > src/CMakeFiles/gpuarray-static.dir/gpuarray_buffer_cuda.c.o > [ 54%] Building C object src/CMakeFiles/gpuarray-static.dir/util/xxhash.c.o > [ 55%] Building C object src/CMakeFiles/gpuarray.dir/util/strb.c.o > [ 57%] Building C object src/CMakeFiles/gpuarray.dir/util/xxhash.c.o > /home/local/ASUAD/rvenka10/libgpuarray/src/gpuarray_blas_cuda_cublas.c: In > function ‘hgemm’: > /home/local/ASUAD/rvenka10/libgpuarray/src/gpuarray_blas_cuda_cublas.c:445:52: > error: ‘CUBLAS_DATA_HALF’ undeclared (first use in this function) > ((uint16_t *)A->ptr) + offA, CUBLAS_DATA_HALF, lda, > ^ > /home/local/ASUAD/rvenka10/libgpuarray/src/gpuarray_blas_cuda_cublas.c:445:52: > note: each undeclared identifier is reported only once for each function it > appears in > /home/local/ASUAD/rvenka10/libgpuarray/src/gpuarray_blas_cuda_cublas.c: In > function ‘hgemm’: > /home/local/ASUAD/rvenka10/libgpuarray/src/gpuarray_blas_cuda_cublas.c:445:52: > error: ‘CUBLAS_DATA_HALF’ undeclared (first use in this function) > ((uint16_t *)A->ptr) + offA, CUBLAS_DATA_HALF, lda, > ^ > /home/local/ASUAD/rvenka10/libgpuarray/src/gpuarray_blas_cuda_cublas.c:445:52: > note: each undeclared identifier is reported only once for each function it > appears in > src/CMakeFiles/gpuarray-static.dir/build.make:381: recipe for target > 'src/CMakeFiles/gpuarray-static.dir/gpuarray_blas_cuda_cublas.c.o' failed > make[2]: *** > [src/CMakeFiles/gpuarray-static.dir/gpuarray_blas_cuda_cublas.c.o] Error 1 > make[2]: *** Waiting for unfinished jobs.... > src/CMakeFiles/gpuarray.dir/build.make:381: recipe for target > 'src/CMakeFiles/gpuarray.dir/gpuarray_blas_cuda_cublas.c.o' failed > make[2]: *** [src/CMakeFiles/gpuarray.dir/gpuarray_blas_cuda_cublas.c.o] > Error 1 > make[2]: *** Waiting for unfinished jobs.... > CMakeFiles/Makefile2:89: recipe for target > 'src/CMakeFiles/gpuarray.dir/all' failed > make[1]: *** [src/CMakeFiles/gpuarray.dir/all] Error 2 > make[1]: *** Waiting for unfinished jobs.... > CMakeFiles/Makefile2:126: recipe for target > 'src/CMakeFiles/gpuarray-static.dir/all' failed > make[1]: *** [src/CMakeFiles/gpuarray-static.dir/all] Error 2 > Makefile:140: recipe for target 'all' failed > make: *** [all] Error 2 > > Here is my nvcc --version output: > nvcc: NVIDIA (R) Cuda compiler driver > Copyright (c) 2005-2016 NVIDIA Corporation > Built on Tue_Jan_10_13:22:03_CST_2017 > Cuda compilation tools, release 8.0, V8.0.61 > > and here is my nvidia-smi output: > Sat Mar 4 02:07:59 2017 > > +-----------------------------------------------------------------------------+ > | NVIDIA-SMI 375.26 Driver Version: 375.26 > | > > |-------------------------------+----------------------+----------------------+ > | GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. > ECC | > | Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute > M. | > > |===============================+======================+======================| > | 0 NVS 310 Off | 0000:01:00.0 N/A | > N/A | > | 30% 48C P0 N/A / N/A | 359MiB / 444MiB | N/A > Default | > > +-------------------------------+----------------------+----------------------+ > | 1 GeForce GTX 1080 Off | 0000:02:00.0 Off | > N/A | > | 27% 37C P8 6W / 180W | 1MiB / 8113MiB | 0% > Default | > > +-------------------------------+----------------------+----------------------+ > | 2 GeForce GTX 1080 Off | 0000:03:00.0 Off | > N/A | > | 27% 37C P8 6W / 180W | 1MiB / 8113MiB | 0% > Default | > > +-------------------------------+----------------------+----------------------+ > > > > +-----------------------------------------------------------------------------+ > | Processes: GPU > Memory | > | GPU PID Type Process name Usage > | > > |=============================================================================| > | 0 Not Supported > | > > +-----------------------------------------------------------------------------+ > > Futher more, with cuda backend, the test_gpu scirpt runs on gpu but only > with about 25% - 30% GPU Utilization. > I use cuDNN 5.0 (not 5.1). This is very persistent and I have tried > everything I can and I am getting nowhere. > > Any help is appreciated. Thank you. > > -- > > --- > You received this message because you are subscribed to the Google Groups > "theano-users" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > For more options, visit https://groups.google.com/d/optout. > -- --- You received this message because you are subscribed to the Google Groups "theano-users" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
