The Concurrent Number Cruncher (CNC) is a high-performance preconditioned conjugate gradient solver on the GPU using the new GPGPU AMD-ATI CTM and NVIDIA CUDA APIs. The CNC is based on a general optimized implementation of sparse matrices using Block Compressed Row Storage (BCRS) blocking strategies for various block sizes, and optimized BLAS operations through massive parallelization, vectorization of the processing and register blocking strategies.