could not use more than one gpu

[AndyTang15]

hi, when I use one gpu, the code works well. But when I change opts.train.gpus = 1; (in cnn_ucf101_temporal.m) to opts.train.gpus = [1,2]; it was fail:

Error using cnn_train_dag (line 120) Error detected on worker 1.

Error in cnn_ucf101_temporal (line 231) [info] = cnn_train_dag(net, imdb, fn, opts.train) ;

Caused by: Error using cnn_train_dag>map_gradients (line 503) Invalid file identifier. Use fopen to generate a valid file identifier.

I've tested opts.train.gpus = [1,2] on the mnisit in Matconvnet example, it works well. anyone could help me?

[JasmineeYang]

Hello,did you run the codes successfully?when I run the code ,such error appears :+1: 错误使用 vl_nnconv Out of memory on device. To view more detail about available memory on the GPU, use 'gpuDevice()'. If the problem persists, reset the GPU by calling 'gpuDevice(1)'. Have you ever met this situation?I want to know where can i change cudnnWorkspaceLimit

[JasmineeYang]

@AndyTang15

[icyzhang0923]

you need to compile the matconvnet again under the condition more than one gpu~try it~

[ml930310]

@JasmineeYang I have a same problem ,do you have solved?Thanks.

[ml930310]

@JasmineeYang

[soumitrasamanta]

@ml930310 I have the same problem when I run the individual networks (spatial or temporal) using multiple GPUs! I found this is due to the existence of folder mentioned in "opts.train.memoryMapFile = fullfile(tempdir, 'ramdisk', ['matconvnet' num2str(feature('getpid')) '.bin']) ;" in both cnn_ucf101_spatial.m & cnn_ucf101_temporal.m files and solved this by creating the folder mentioned in "opts.train.memoryMapFile".

[ml930310]

sorry,I haven't done this for a long time, I forgot it.

At 2018-08-10 22:42:54, "SS" notifications@github.com wrote:

@ml930310 I have the same problem when I run the individual networks (spatial or temporal) using multiple GPUs! I found this is due to the existence of folder mentioned in "opts.train.memoryMapFile = fullfile(tempdir, 'ramdisk', ['matconvnet' num2str(feature('getpid')) '.bin']) ;" in both cnn_ucf101_spatial.m & cnn_ucf101_temporal.m files and solved this by addition adding one line which create the folder mentioned in "opts.train.memoryMapFile".

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.