I am trying to run a program to train by CNN on Matlab. However, I realize that this CNN-trained program requires MathConvNet library. This library requires GPU-supporter. Are there any ways to run CNN training on unsupported-GPU computer?
If for MATLAB is impossible, Can I try to train CNN on other platforms, such as C++, Java, Python etc. without GPU?
Note:
1. My MATLAB version 2018a.
2. My configuration: Intel core i5 (Haswell), 4GB RAM, The built-in Intel Graphics with 1GB memory (integrated GPU, not separate GPU).
2. When I try to run gpuDevice on Matlab, I got the following message:
Error using gpuDevice (line 26)
There is a problem with the graphics driver or with this GPU device. Be sure that you have a supported GPU and that the latest driver is installed.
Caused by:
The CUDA driver could not be loaded. The library name used was 'nvcuda.dll'. The error was:
The specified module could not be found.
or if I run gpuDeviceCount, it returns 0.
Dears Dang,
For using CNN without GPU supported, you can use MatConvNet.
Please site this URL : http://www.vlfeat.org/matconvnet/install/
CNN can be run by other programming language such python and java.
If you use python, you can use some optional framework such : Torch, Tensorflow, Theano and etc. For Java programming language, usually use deeplearning4j (https://deeplearning4j.org/) framework.
Hope it will help you.
Regards
Toto
Dears Dang,
For using CNN without GPU supported, you can use MatConvNet.
Please site this URL : http://www.vlfeat.org/matconvnet/install/
CNN can be run by other programming language such python and java.
If you use python, you can use some optional framework such : Torch, Tensorflow, Theano and etc. For Java programming language, usually use deeplearning4j (https://deeplearning4j.org/) framework.
Hope it will help you.
Regards
Toto
https://uk.mathworks.com/matlabcentral/answers/284601-convnn-train-in-cpu-mode
https://www.researchgate.net/post/How_to_use_MATLAB_deep_learning_without_GPU
Dear Thanh,
In python, by using TensorFlow/Keras frameworks you can use Deep Learning approaches (including CNN) using CPU only. However, the training/testing stage are computationally inefficient (regarding using GPU)
@Artur: I knew that training on GPU has higher performance. The problem is that my computer just has integrated GPU, not separate GPU. I am not ready to order new computer now :)
@Toto: thank you for that info. I searched on Mathworld, there is a topic, in which the answer is the MathConvNet is only support on GPU :) That is wrong! The link you sent is helpful. Thank you!
In Matlab 2018a, you can first define the architecture CNNs and then train it by not installing any external library . By default, it trains the defined architecture on CPU.
https://www.mathworks.com/help/deeplearning/examples/create-simple-deep-learning-network-for-classification.html
To train the defined architecture on GPU, you first need to configure it first, which is also very trivial.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Programming Languages
- Python