ERROR: too many resources requested for launch img2col when use pytorch

Robotics & Edge Computing Jetson & Embedded Systems Jetson TX2
1

I’m using pytorch on a tx2 with jetpack3.2.
The pytorch (0.3.0) is compiled from source as https://github.com/dusty-nv/jetson-reinforcement/blob/master/CMakePreBuild.sh.

I use it to do a forward computing. In my PC with a 1080ti, it needs 3GB memory to do this forward computing. But in tx2, it always gives me this “too many resources requested error” when the torch module do forward computing.

That’s why?

2

Hi,

Could you reboot the device and try it again?
Thanks.

3

I’m running into the same error when I try to run a pretrained PyTorch network on some imagenet images.
I compiled PyTorch v0.4.0 from Source and run

python examples/imagenet_eval.py ../../datasets/imagenet -a vgg16 -b 20 -e

at the root of this repo: GitHub - Cadene/pretrained-models.pytorch: Pretrained ConvNets for pytorch: NASNet, ResNeXt, ResNet, InceptionV4, InceptionResnetV2, Xception, DPN, etc.
Setting number of workers to zero or one doesn’t help it.

I also encountered memory issues on other PyTorch examples (SIGKILL (n_workers = 0) or “Connection Refused” (n_workers > 0) on worker shutdown in DataLoader).

Any ideas?

4

Hi,

Could you set batchsize=1 and try it again?

python examples/imagenet_eval.py ../../datasets/imagenet -a vgg16 -b 1 -e

Thanks.

5

Hi AastaLLL,
thank you for your reply.
For batchsize=1 it results in the same error.

6

Hi,

Could you monitor the memory utilization with tegrastats and share data with us?

sudo ./tegrastats

Thanks.