Hello- I installed the NVidia DIGITS Amazon AMI image, running g2.2xlarge instance. I set up the ‘object-detection’ example (from the examples folder in the DIGITS github project) and started training based on the downloaded KITTI dataset according to the ‘object-detection’ instructions.
It is training extremely slowly. Estimated completion in 4+ days! Only did 800 iterations in 45 minutes. The instance has 15G memory (on CPU I think?) and 4G on GPU (according to digits/device_query.py). This is much slower than it would probably be running on my own GPU, but I can’t use that for various reasons. Any idea why so slow?