What computer configuration should I choose for machine learning?

Question

Machine learning

Greetings.
Started studying machine learning. I decided to train a small neuron on text files 2 x 20 meg, and I ran out of memory in the process - the word vectors did not fit)
Now I have an old i7 8GB of RAM, and nvidia 560
Apparently I need to upgrade the computer - what to take?
New i7, 32gb ram and nvidia 1080ti?
What configuration to choose?

Answer 1

I'd rather optimize for memory at the expense of speed. And there will be a dataset for 100 megabytes - will you buy 64 gigabytes?
Try to form vectors only for each minibatch, and when the minibatch passes, release the memory. That is, not immediately translate everything into vectors, but in the process of work. Yes - longer. Yes, you will repeat everything anew in every era. But memory efficiency has an inverse relationship with speed efficiency...
The config is quite normal, despite the fact that in 560 CUDA version 2.1, the same TensorFlow is not supported for acceleration (where the minimum is in the current version 3.0). At the most this vidyuha on one of the computers. Therefore, if you want to use CUDA, then I advise you to buy a new vidyuhi, and with memory, always preprocess separately for the minibatch, and not for the entire dataset.
PS Config 1 in 1 like my second computer))))