search

faustomilletari / VNet

GNU General Public License v3.0
284 stars 123 forks source link

What GPU did you use? #6

Closed ljpadam closed 7 years ago

ljpadam commented 7 years ago

What GPU did you use? I have a Titan x GPU with 12g video memory, but it can not load the "train_noPooling_ResNet_cinque.prototxt" as the input data is 12812864.

faustomilletari commented 7 years ago

Are you using cudnn? I have worked with gtx1080 which has just 8gb. Be sure you are working with 3dcaffe which is in another repository of mine...

Fausto Milletarì Sent from my iPhone

On 10 Nov 2016, at 07:17, ljpadam notifications@github.com wrote:

What GPU did you use? I have a Titan x GPU with 12g video memory, but it can not load the "train_noPooling_ResNet_cinque.prototxt" as the input data is 12812864.

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

ljpadam commented 7 years ago

Thank you for you reply. I used your 3dcaffe and cudnn, but I can load the data with the dimension of 1 * 1 * 64 * 64 * 32 at most. And what the version of cudnn do you use? When you load "train_noPooling_ResNet_cinque.prototxt", how much video memory did it consume?

faustomilletari commented 7 years ago

With the provided code it should run in less than 8 gb and batch size 2 and resolution 128 128 64.

Fausto Milletarì Sent from my iPhone

On 10 Nov 2016, at 08:53, ljpadam notifications@github.com wrote:

Thank you for you reply. I used your 3dcaffe and cudnn, but I can load the data with the dimension of 11646432 at most. And what the version of cudnn do you use? When you load "train_noPooling_ResNet_cinque.prototxt", how much video memory did it consume?

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

ljpadam commented 7 years ago

Sorry to disturb you again. And which version of cudnn do you use? When I complied your 3dcaffe, some incompatible problems of cudnn occured. Currently, it can only be successfully compiled with the cudnn on my machine after I modified some codes related to cudnn of your 3dcaffe. I don't know whether this causes consuming more memory.

faustomilletari commented 7 years ago

It could be. I used cudnn 5 (the latest) the out of memory problem happened to other people who were not using cudnn.

Fausto Milletarì Sent from my iPhone

On 10 Nov 2016, at 09:22, ljpadam notifications@github.com wrote:

Sorry to disturb you again. And which version of cudnn do you use? When I complied your 3dcaffe, some incompatible problems of cudnn occured. Currently, it can only be successfully compiled with the cudnn on my machine after I modified some codes related to cudnn of your 3dcaffe. I don't know whether this causes consuming more memory.

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

faustomilletari commented 7 years ago

please refer to this as well https://github.com/faustomilletari/VNet/issues/2