Pytorch using 90+% ram and cpu while having GPU

(Shahariar Rabby) #1

While training the resnet from lesson 1 taking lots of time (15min+).
I tried all possible solutions from the forum and nothing helped.
My device Asus x556 with 940mx GPU, Core i5, 8 gig Ram and 250gb SSD with windows 10.

When I use tensorflow it always uses 90% GPU and no ram or CPU uses. Any way to do the same for the torch?


(Aditya) #2

It’s using the GPU…
Also 940mx isn’t suitable…(I also have it so I know, The real GPU series starts from 1050+…)

(Shahariar Rabby) #3

Is there any way to stop using CPUa and RAM? I can’t use the PC while training. :frowning:

(Shahariar Rabby) #4

It’s also giving memory error :frowning:

(Aditya) #5

The only way out is not to train on your laptop…
Just understand things properly and train them on a powerful machine or use AWS…

Mem error is because you are trying to allocate more memory than you have…


I have the same problem, but with a GTX 1060. My CGU is well recognized:

torch.cuda.current_device() -> 0
torch.cuda.get_device_name(0) -> 'GeForce GTX 1060'

But when computing, cpu is widely used over cgu:

And it takes several minutes while I expect it to take several seconds.

How can I be sure that the cgu is used?


I encountered a similar issue using a GTX 1080 ti GPU on windows 10.
CPU: Intel® Core™ i7-6700 CPU @ 3.40GHz
RAM: 16384MB RAM
Has anyone found a solution?

(kushal agrawal) #8

How to make the training run on GPU ??
i have install cuda and cudnn but i m not able to proceed further.