GPU memory on AWS instance

I have a g2.2xlarge instance (http://ift.tt/1hHKqUH) on AWS running Ubuntu 10.4 with GTX 750 Ti GPU running cuda. The GPU has 4GB of memory. I am using the instance to run Theano, (http://ift.tt/OaLIE2), a popular machine learning library. My program aborts with following error

Error allocating 1073741824 bytes of device memory (out of memory). 
Driver report 888422400 bytes free and 4294770688 bytes total

The program requests about 1GB of GPU memory while only 800MB is free. How can I know what processes are using GPU memory? I am not running any other process on the instance. I tried looking into nvidia-smi (http://ift.tt/1gNsGVN), but could not get any information.