Send your promotional news to: email@example.com
The GPU has its own device memory, and is transferred between the GPU and host memories using programmed direct memory access, which operates concurrently between both the CPU and GPU. The device memory supports a very high memory bandwidth through the use of a wide data path. However, since the GPU is a coprocessor, usually on a separate bus device, data must first be copied explicitly from the CPU to device memory on the GPU. This can give rise to performance bottlenecks. In a recent paper by author Dr. P. M. Darbyshire, clever use of memory optimisation provided a dramatic performance improvement and elevated some of the bottleneck issues.
"Not only is it important to have a detailed knowledge of the parallel programming model but also an understanding of how the hardware architecture can help improve and overcome performance issues to push even more compute power out of these devices". Indeed, such knowledge will greatly help in tackling more complex mathematical models allowing us to fully exploit the advantages of using parallel processing on high performance computers for cancer research", Darbyshire said.
In the last decade in silico experiments focused on tumour growth have become more readily accepted by the biological community both as a means to direct new research and a route to integrate multiple experimental measurements in order to generate new hypotheses and testable predictions. More recently, the advantages of using parallel processing has highlighted the potential gained from the numerical solution of complex mathematical models using such computational techniques. Indeed, it is assumed that the implementation of more advanced cancer models that can be handled in a massively parallel manner will provide an extremely useful virtual experimental platform. With such tools, clinicians and oncologists will have the ability to develop new treatment scenarios and simulate these in a realistic time frame, which will substantially reduce pre-clinical trial costs and associated time constraints.
Dr P. M. Darbyshire, Technical Director, Department of Computational Biophysics, Algenet Cancer Research, Nottingham. UK.
A Hybrid Continuous-Discrete Model of Tumour-Induced Angiogenesis is Solved Numerically in Parallel and Performance Improvements Analysed. doi: 10.11648/j.ejb.20150305.11.