Marvin is a deep learning framework designed first and foremost to be hackable. It is naively simple for fast prototyping, uses only basic C/C++, and only calls CUDA and cuDNN as dependencies.
Modern graphics processing units (GPUs) contain hundreds of arithmetic units and can be harnessed to provide tremendous acceleration for many numerically intensive scientific applications. The key to effective utilization of GPUs for scientific computing
Harte Zahlen zur Rechenleistung künftiger Nvidia-GPUs gab es jedoch nicht, die Einheit, in der Huangs Roadmap auf der Vertikalen skaliert ist "GFlops pro Watt". Wie der Nvidia-Mitbegründer betonte, ist die Rechenleistung nicht das Problem, sondern die "Power Wall". Schon mit den ersten Fermi-Grafikkarten kratzte Nvidia an der Grenze von 300 Watt.
A. Cheik Ahamed, and F. Magoulès. Distributed Computing and Applications to Business, Engineering and Science (DCABES), 2014 13th International Symposium on, page 19-23. (November 2014)
A. Cheik Ahamed, and F. Magoulès. High Performance Computing and Communications, 2014 IEEE 6th Intl Symp on Cyberspace Safety and Security, 2014 IEEE 11th Intl Conf on Embedded Software and Syst (HPCC,CSS,ICESS), 2014 IEEE Intl Conf on, page 121-128. (August 2014)
J. Choi, A. Singh, and R. Vuduc. Proceedings of the 15th ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming, page 115--126. New York, NY, USA, ACM, (2010)