The paper discusses the implementation and performance of the PARDISO solver for solving unsymmetric sparse systems of linear equations. The authors introduce a supernode pivoting approach, which, combined with supernode partitioning and asynchronous computation, achieves high performance on shared memory parallel computers. The method involves prepermuting rows to maximize the magnitude of diagonal entries and using static pivoting to reduce the amount of pivoting during factorization. The BLAS-3 level efficiency is maintained, and an enhanced left-right looking scheduling scheme is used to improve speedup without increasing operation count. Experiments demonstrate that the method can solve a wide range of unsymmetric linear systems from real-world applications, achieving high performance on large sparse matrices. The paper also compares PARDISO with other well-known software packages, highlighting its stability and scalability. Future improvements include better reordering schemes, dynamic pivoting, and postprocessing techniques to further enhance robustness and efficiency.The paper discusses the implementation and performance of the PARDISO solver for solving unsymmetric sparse systems of linear equations. The authors introduce a supernode pivoting approach, which, combined with supernode partitioning and asynchronous computation, achieves high performance on shared memory parallel computers. The method involves prepermuting rows to maximize the magnitude of diagonal entries and using static pivoting to reduce the amount of pivoting during factorization. The BLAS-3 level efficiency is maintained, and an enhanced left-right looking scheduling scheme is used to improve speedup without increasing operation count. Experiments demonstrate that the method can solve a wide range of unsymmetric linear systems from real-world applications, achieving high performance on large sparse matrices. The paper also compares PARDISO with other well-known software packages, highlighting its stability and scalability. Future improvements include better reordering schemes, dynamic pivoting, and postprocessing techniques to further enhance robustness and efficiency.