I found a great blog post here: http://ankivil.com/installing-keras-theano-and-dependencies-on-windows-10/ The issue is that it doesn't specify linking cuDNN. I was wondering if anybody knew how to get this to work properly. Does one simply install cuDNN and magic happens or are there further steps?
Thanks!
Actually to linking CuDNN you just need to put 4 files that you've downloaded from https://developer.nvidia.com/cudnn.
the directory files you'll get are:
cuda
|-bin
|-include
|-lib
copy files from those directory to the approppiate path, in my windows 10 system at C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v7.5/
Related
Like I said before, after installing TensorFlow-gpu by anaconda im getting this error in python terminal
W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'cudart64_100.dll'; dlerror: cudart64_100.dll not found
The thing is, that this file actually exists in this dir (maybe this will help) C:\Users\MyUser\Anaconda3\envs\Gpu_Tensor_Flow\Library\bin
Please help me, because I have no idea what I've done wrong.
I got the same issue
Jozef Jarosciak published an answer that worked for me, visit https://www.joe0.com/2019/10/19/how-resolve-tensorflow-2-0-error-could-not-load-dynamic-library-cudart64_100-dll-dlerror-cudart64_100-dll-not-found/
I downloaded the cudart64_100.dll directly from his website and put it in this path (using windows, i guess it is the same for you since install on linux is pretty easy) C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\bin\
It is weird because I installed all software requirements with lastest version, still tensorflow seems to load this old dll, may be tensorflow is not up to date yet.
The Graphviz download page has to links, one for a .msi file and another for a .zip
when running the .msi, it looks that it's for 32 bits, since the default directory installation is "Program Files (x86)", which is where the 32-bit installations go.
There are no instructions on what to do with the .zip
Has anyone installed the 64-bit version of Graphviz? Is there an installation script that installs the executables? If so, where can I find it?
(I hope it exists, because I can't compile source code in my machine.)
Many thanks in advance,
Alex
A litte late, but in case someone needs an answer : By now, there seems to be no official 64bit support by graphviz.
I found a discussion on github about graphviz for python where you can find a link to a github repository, that leading to an adapted graphviz solution for Visual Studio with 64bit support.
In another discussion on github a link to an AppVeyor project is provided, which also includes a 64bit configuration. Unfortunately this one seems to be abandoned.
I try to install MinGW-w64, but during install i got
the file has been downloaded incorrectly
I have 64 bit win10 and my option of installation:
I use that link sourceforge. and some commentators have same problem.
Any idea how to fix it?
Newer MinGW-w64 releases are now available at https://winlibs.com/ as standalone packages that don't need an installer. Just extract to a folder and start using. This also allows having multiple versions on your side by side (e.g. 32-bit and 64-bit).
This MinGW article addresses the issue. It's actually pretty simple to fix. The problem is with the Windows installer. What you need to do is download the MinGW zip archive directly (not the installer) and extract the files to something like C:\mingw and then add it to PATH in Environment Variables. Here's direct link to the SourceForge files you'll need.
That's increasingly frustrating.. I'm trying to build OpenCV for Java on a Raspberry Pi and that requires JNI.
When executing Cmake, it finds ANT correctly but it always says:
Could NOT find JNI (missing: JAVA_INCLUDE_PATH JAVA_INCLUDE_PATH2 JAVA_AWT_INCLUDE_PATH)
By now I have tried at least a dozen suggestions from the web but without any success. Even tried the modified cmake file like here: https://gist.github.com/emchristiansen/6845954
The JAVA_HOME path is also correct as ls $JAVA_HOME leads me to the folder of Oracle's Java 8 jdk and there are the JNI files in a subfolder...
A global search finds them too.
What could be other reasons for this problem?
Big thanks in advance.
Or on a side question.. does someone happen to have a pre-build version of OpenCV from a Raspberry the could offer for download?
I only need the native library (.so) and .jar file. Would be really great!
I'm trying to install the cuda-accelerated library found on this website : http://mklab.iti.gr/project/GPU-LIBSVM
I downloaded the linux makefile and executable and tried to follow the readme file :
Instructions to compile Linux GPU-Accelerated LIBSVM
Install the NVIDIA drivers, CUDA toolkit and GPU Computing SDK code samples. You can find them in:
http://developer.nvidia.com/object/cuda_3_2_downloads.html (January
2011)
You may need some additional packets to be installed in order to
complete the installations above. Please refer to the web for more
details.
Copy this folder to "/NVIDIA_GPU_Computing_SDK/C/src"
Use the Makefile found in "/NVIDIA_GPU_Computing_SDK/C"
Find the "svm-train-gpu" executable in /NVIDIA_GPU_Computing_SDK/C/bin/linux/release
But i could not find the "NVIDIA_GPU_Computing_SDK" folder, even after upgrading cuda to 5.5 (wich should contain the SDK). It seems that this instructions are outdated. So I copied the folder into "opt/cuda/src/" and used this makefile :
EXECUTABLE := svm-train-gpu
CUFILES :=
CU_DEPS :=
CCFILES := svm-train.c svm.cpp
CUDACCFLAGS := -po maxrregcount=16
USECUBLAS := 1
include ../../common/common.mk
It first told me it could not find "common.mk", so I found in my disk an other common.mk file from an other CUDA project implementation called "common-gcc-cuda-4.0.mk" wich is described in the header as a "Common build script for CUDA source projects for Linux and Mac platforms" and tried again.
Now I have this error :
$ make
make: *** No rule to make target `makedirectories', needed by `bin/linux/release/svm-train-gpu'. Stop.
Can anyone help me please ?
It sounds like the library is expecting an old version of CUDA. The structure of the sample codes (GPU_Computing_SDK) changed substantially after CUDA 4.2
Try starting over with an older version of CUDA which you can find here. I would recommend choosing a cuda version from 3.2 to 4.2, and if you install that completely (driver, toolkit and SDK) you will find the folders and makefiles being referenced.
The version you were trying was created with the CUDA SDK 3.0. As Robert pointed out the structure of sample codes and makefiles changed substantially after CUDA 4.2.
A new version (1.2) is online on the mklab website http://mklab.iti.gr/project/GPU-LIBSVM
which supports the latest LIBSVM v3.17 and CUDA SDK v5.5.
Also you can see a detailed changelog here: https://github.com/MKLab-ITI/CUDA/releases