We propose a drip-feeding cooling system that efficiently cools high thermal density servers using multiprocessor implementation and General-purpose computing on graphics processing units (GPGPU) used for high-performance computing such as machine learning. We evaluated this system through computational fluid dynamics (CFD) simulation and an experiment. In the CFD simulation, two types of fluorine inert liquid and three types of silicone oil were used as refrigerant. In the experiment, one type of fluorinated inert liquid and one type of silicone oil were used. As a result, this system could efficiently cool heat generation by about 16 kW per rack at a power usage efficiency of 1.04 or less and reduce a floor load to 500 kg/m² (102 lbs/ft²) or less.