Hello folks, I am struggling the last few weeks in training large DNNs in my local machine due to poor CPU/GPU performance. Does anyone know if there is any cloud service that we can use for training machine learning algorithms that require great computational power? Something that we can just upload the code and data and hit Run.
I think you left out the and provide credit card part
Both Google and AWS allow you to buy time and resources if their free offerings are insufficient for your needs.
The last time I looked, which admittedly was maybe a year ago, Google Colab can be used in “free” mode, as ai_curious points out. The only problem is that you are not guaranteed access to a GPU/TPU at any particular time: you may have to wait in a queue if the paying users are consuming all the resources. But if there is resource available, they let you use it for free. I did not look into what the billing model is, but I had no trouble running model training tasks for free. If you need to run for seriously long times, then there are a few other hassles you have to deal with, e.g. in non-paying mode, they suspend your session if there is no keystroke activity for a while, but there are “hacks” you can implement to get around that.
AWS also provides services like this, but when I used those through work a few years back there was no equivalent “free” model as I described above for Colab. But there again my information may not be current. I think AWS has a website with this information.
There is also a throttle on GPU. When taking the TensorFlow AT Specialization just last month I got kicked off for 24 hrs after hitting it. The OP asked for free access to great computational power.. If there is a such a thing I want to know so I can short their stock.
Fair enough! Maybe the better way to state it is that you can definitely get greater computational power than your laptop for free. Whether that counts as “great power” is in the eye of the beholder, I guess. As usual, there is no free lunch for the reasons you allude to.
Great, thanks guys! I will look into the Google and AWS services!