New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
compile the caffe #19
Comments
Sorry but we modified the official caffe for our project. So we rely on openmpi if you want to use multiple GPUs (we only use multi-gpu for testing, but not training). We also highly recommend you install cudnn v5. After downloading and extracting it, you need to replace the cmake .. -DUSE_MPI=ON -DCUDNN_INCLUDE=/usr/local/cuda/include -DCUDNN_LIBRARY=/usr/local/cuda/lib64/libcudnn.so |
Thanks,But according to my environment which has only one server with 4 GPUs,can I use the openmpi? |
Sure. You can change these two lines to mpirun -n 4 python2 tools/eval_test.py \
--gpu 0,1,2,3 \ |
um,thanks. |
Yes. It should be >= 1.55. |
A requested component was not found, or was unable to be opened. This Host: amax-1080
|
It seems that you have different versions of openmpi. Let's say if you compile openmpi and install it into a local directory like export PATH=/home/jxd/openmpi/bin:$PATH
export LD_LIBRARY_PATH=/home/jxd/openmpi/lib:$LD_LIBRARY_PATH Restart the terminal, |
Hello,I have successfully installed the openmpi,and test it that it can be used.Then I cmake the caffe successfully,but I still exist the questions above.So I try to do the training,it meets the same questions.
A requested component was not found, or was unable to be opened. This Host: amax-1080
|
Could you please check the output of the following commands: which mpirun
ldd $(which mpirun) | grep mpi
ldd caffe/build/install/bin/caffe | grep mpi |
yeah,maybe I do not cmake caffe successfully as there's no information about it ?
jxd@amax-1080: |
OK. You have another self-compiled openmpi installed at export PATH=/usr/local/openmpi/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/openmpi/lib:$LD_LIBRARY_PATH Restart the terminal, remove the build directory under caffe, and recompile it following the steps in the README file. |
Yes,I have added these lines to ~/.bashrc,and recompile it yesterday.Are there two openmpi installed in the system? |
Right. In your previous log, it complaints mca: base: component_find: unable to open /usr/lib/openmpi/lib/openmpi/mca_paffinity_hwloc: perhaps a missing symbol, or compiled for a different version of Open MPI? (ignored) So you have a system-installed openmpi at |
thanks a lot!
all detection: |
Good to hear that! Will close the issue for now, and please feel free to reopen it if there are further problems. |
hello,when I run the make -j8 && make install,it shows the error following:
[ 87%] [ 88%] make[2]: *** No rule to make target
/path/to/cudnn/lib64/libcudnn.so', needed by
lib/libcaffe.so'. Stop.make[2]: *** Waiting for unfinished jobs....
Building CXX object src/caffe/CMakeFiles/caffe.dir/data_transformer.cpp.o
Building CXX object src/caffe/CMakeFiles/caffe.dir/syncedmem.cpp.o
make[1]: *** [src/caffe/CMakeFiles/caffe.dir/all] Error 2
make: *** [all] Error 2
I wonder if it is the error of the path?
Another question:
Can I unuse the "cudnn"and "openmpi"?I only have one server but have 4 gpus,Iwonder if I can use the "-gpu al" to replace the "openmpi".
thanks!
The text was updated successfully, but these errors were encountered: