If you need to spawn 10-100+ instances for heavy computation tasks - then AWS and alikes are the only choice. @$2k/mo+/instance - for smaller scale research it

If you need to spawn 10-100+ instances for heavy computation tasks - then AWS and alikes are the only choice. @$2k/mo+/instance - for smaller scale research it is way cheaper to buy high performance GPU card(s) yourself and use them without unpredictable recurring cost issues. Computers are cheap. https://www.reddit.com/r/MachineLearning/comments/45vepw/choose_a_good_cuda_card_looking_for_your_advice/ https://www.reddit.com/r/MachineLearning/comments/45vepw/cho... http://www.videocardbenchmark.net/high_end_gpus.html http://www.videocardbenchmark.net/high_end_gpus.html