Unlike a traditional GPU, Nvidia's first general purpose GPU series, Tesla, does not output videos to a display. It is developed for high performance computing. In this project, I will be using CUDA enabled Tesla C1060 GPUs to speed up nbody simulations. Each C1060 GPU has 240 cores. We expect the simulation to run at least ten times faster compare to a typical workstation. The GPU server is purchased from Colfax International. The base platform is manufactured by Super Micro. It has two Tesla C1060 GPUs, two Intel Xeon Quad Core E5504 CPUs, and six 2GB RAMs.
The CUDA machine was compared with one node on Jeremy's cluster, JMD.
|CPU||2 X Intel Xeon E5504||AMD Opteron 844|
|GPU||2 X Nvidia Tesla C1060||N / A|
|RAM||6 X 2GB||2GB|