I mean all possible work with tagging data: GIS, tagging data pre-processing, visualisation, different types of modelling and modern analysis by using R, Python, etc.
I'm assuming that by "hardware" you mean PC hardware. Most data science libraries, especially those for R are optimized for Intel processors, although more and more these days models are run in parallel using GPUs. GPUs are phenomenally good at doing the kinds of repeated simple calculations required for most complex machine learning models but the model needs to be specifically written in order to make use of a GPU. nVidia GPUs are currently the leaders due to their CUDA framework.
If you want a good setup that will enable you to do most of what you'll need to do short of the kinds of things that would be best left to cloud computing, I'd recommend a beefy Intel workstation processor and as good an nVidia graphics card (preferably in the Quadro line, though regular gaming cards work well too) as you can afford. Don't skimp on the RAM, as some workloads will require loading the whole dataset into RAM for processing and GIS datasets (rasters, etc.) in particular can get pretty hefty.
I second David's suggestion. Just to add: If you have really large data sets (TB) and or you want to run really complex simulations which may be parallelized, you might consider using a computer cluster. Universities often have such clusters, and there are possibilities to by access to commercial clusters. It saves you own investment and administration.
Thanks to all! Yes, I meant PC hardware and edited my question.
So, what if I buy the latest intel processor and 32 gigabytes of RAM for working with tabular data (I think it would be 30-40 thousand locations to work with after initial filtration)? Am I right that there is no strong need for a graphics card when working with such tables?
I am only studying this area, so I apologize for such questions.
It depends a bit on what you mean with "working with". If this is just for visualization or standard statistical analyses, any reasonably recent cheap PC will do. If you work on complex simulations, you need processor power (and RAM). It might be that using 6-12 cores in parallel is sufficient, but depending on the task it may be advantagous to think about using GPUS or even clusters.