In this talk, we study the performance of regularized least squares (RLS) with box relaxation as it applies to massive MIMO. We derive a precise BER  expression at high dimensions using the Convex Gaussian Min-max theorem, a recently developed tool that emerged from the theory of compressive sensing. We take advantage of the accurate performance prediction results to optimally tune the relevant parameters, namely the regularizer and the box threshold which will be shown to be SNR dependent. We finally show how to accurately estimate the SNR from a single observation.