But if you want more computing power you pay for powerups right?
but the powerups just give you longer access to gpu NOT a better processor, right?
What do you mean with a better processor? What is the term of comparison?
I have ran some programs on the free tier GPU and its about the same as my desktop CPU.
If you have a quad-core CPU, the performance is pretty much the same.
How can i improve my processing power as cheaply as possible? without paying several dollars per hour?
There isn't a clear answer because it's depending on your needs(the type of your experiments, your meaning of cheap and so on). I can give you a more general advice about how setup a good workflow:
- Sketch the code locally or on in interactive mode(
--mode jupyter) in a CPU instance
- Iterates on CPU instance to fix the code and make your script compliant with the FloydHub workflow and environment
- Switch to GPU for the training, before that, make sure to take advantage of the full hardware
- If you want to run faster, Switch to GPU2 and optimize your deep learning model
Some months ago, we have run some benchmarks and provide some advice about how to optimize your model, here's the page.