From: Sumit Gupta
Date: Dec 5, 2009
GPUs (Graphical Processing Units) are general purpose processors and offer over 10x the performance of a CPU. The model
is of co-processing - you only send the compute intensive code to the GPU, the rest of the code
remains unchanged: www.nvidia.com/object/GPU_Computing.html
GPUs are fully programmable in C, C++
There is some early work on data mining/analytics using GPUs, such as
Data Analytics Software using CUDA
- GPUMiner: Parallel Data Mining on NVIDIA GPUs, Fang et al, Hong Kong University of Science and Technology and Microsoft
- Cmatch: Fast Exact String Matching on the GPU, Schatz, Trapnel, University of Maryland
NVIDIA just announced the Fermi architecture, which offers about 1.3 TF single precision
performance and ~650 double precision : www.nvidia.com/fermi
To find out more, contact Sumit Gupta, Sr. Manager - Tesla GPU HPC Group, NVIDIA
sugupta at nvidia dot com
|