It is a common problem to run (non-linear) SVMs (Support Vector Machines) on large datasets without running into performance issues. Since this problem is caused by complexity, it can’t really be solved. However, there are other implementations than libsvm that speed up the process a bit by using approximations. Further it is questionable, if we really need SVMs anymore since (wide and deep) neural networks work really well now and can be adapted for fast inference as well. Anyhow, here are other SVM packages I found so far:

scikit-learn alternatives to SVR and SVC: