Аннотация:The paper presents capabilities and implementation details for the newly developed XAMG library for solving systems of linear algebraic equations with multiple right-hand sides. The underlying code design principles and the basic data objects implemented in the library are described. Several specific optimizations providing significant speedup compared to alternative state of the art open-source libraries are highlighted. A great attention is paid to the XAMG library thorough performance investigation. The step-by-step evaluation, performed for two compute systems, compares the single right-hand side calculations against hypre, the performance gain due to simultaneous solution of system with multiple right-hand sides, the effect of mixed-precision calculations, and the advantages of the hierarchical MPI+POSIX shared memory hybrid programming model. The obtained results demonstrate more than twofold speedup for the XAMG library against hypre for the equal numerical method configurations. The solution of systems with multiple right-hand sides provides 2–2.5 times speedup compared to multiple solutions of systems with a single right-hand side.