Posted on Thursday, April 30 2009 @ 1:50 CEST by Thomas De Maesschalck
NVIDIA announced Chinese geophysical services provider GeoStar has unveiled a new NVIDIA Tesla based hardware and software solution for oil and gas companies in China. The companies claim the new GPU-based solution is much faster than traditional CPU clusters.
Leveraging the processing power of NVIDIA(R) Tesla(TM) GPUs, GeoStar's seismic software suite dramatically accelerates the performance of complex seismic data. Now, the computation of large datasets generated by searching the earth for oil deposits, can be achieved in smaller, more power efficient GPU based systems as compared to CPU-only based clusters.
"We are dealing with large prestack time migration datasets, which typically take more than 30 hours to run on a cluster of 66 CPUs," said Liu Qin, general manager of GeoStar. "Just a single Tesla C1060 GPU delivers roughly the same computing power, which means we can get orders of magnitude performance increases as we add more GPUs, while dramatically saving power and cost. Tesla GPUs are truly a revolutionary solution for oil and gas exploration."
The Institute of Geology and Geophysics at the Chinese Academy of Sciences (CAS) has spent recent months testing NVIDIA Tesla GPUs with GeoStar's solution. In a computation of prestack time migration data covering 740 square kilometers, 24 Tesla GPUs completed the processing more than 600 times(i) faster than a traditional cluster with 66 CPUs, a result which CAS researchers believe can be improved through additional tuning.
Based on NVIDIA's massively parallel CUDA architecture, NVIDIA Tesla GPU Computing solutions are transforming a broad selection of industries but their impact has been profound in the oil and gas space. As computational needs increase, the power consumption required to run and cool servers in datacenters have become a major portion of operational expenses. Tesla GPU-based clusters enable much higher performance than CPU-only clusters, meaning oil and gas companies can now deploy smaller, more computationally dense clusters that use less energy and still meet the ever increasing demand of applications such as seismic processing.
