We present the results of our lattice numerical calculation of light hadron masses done on a 24-node partition of the VPP500/30 vector-parallel supercomputer at RIKEN. It is the largest of this kind attempted so fa in the history of lattice QCD, using a \(64 \times 48^3\) lattice and the coupling at \(\beta = 6.5\) of a spatial volume of about \((\mbox{2.4 fm})^3\) and lattice cutoff \(a^{-1}\) of about 4 GeV. It allows us to study the region of realistic light-quark masses such as \(m_q \sim \mbox{5 MeV}\) without extrapolation. Metropolis/over-relaxation method is used for the Monte Carlo part and the conjugate gradient method is used for inverting Dirac matrix, both running at more than 25 Gflops sustained. In particular, masses of various light hadrons such as \(\pi, \rho, a_1, b_1, N\) and \(\Delta\) are discussed. We pay close attention to how the systematic errors involved, such as finite physical volume effect, finite lattice spacing effect and finite quark mass effect, are controlled. We also discuss the ``quenched chiral log'' problem which is associated with removing part of the dynamics.