Fine Arts : 215
Date & Time
February 11, 2015, 11:00 am – 12:00 pm
|Session Chair||Hye Kyung Park|
Speaker 1: Samuel Khuvis
- Long-Time Simulations on Modern CPUs and Many-Core Processors
- Modern architectures with multiple memory hierarchies in CPU and coprocessors such as GPU and Intel Phi offer opportunities to drastically speed up numerical kernels. Algorithms that are highly parallel and scale well to over 100 threads can take advantage of the Phi's high degree of parallelism. Simulating calcium waves in a heart cell requires both fine meshes and large final times to match laboratory experiments. This is a perfect example of a problem that can profit from offloading numerical kernels to accelerators such as the 60-core Intel Phi processor on each hybrid compute node, combined with pooling memory from several nodes to allow for the desired spatial resolutions. All results were produced using the maya cluster of the UMBC High Performance Computing Facility.
Speaker 2: Xuan Huang
- Simulation of Calcium Induced Calcium Release in a Heart Cell on Hybrid CPU/GPU cluster
- A mathematical model of Calcium Induced Calcium Release in a heart cell has been developed that consists of three coupled non-linear advection-diffusion-reaction equations. A program in C with MPI based on matrix-free Newton-Krylov method gives very good scalability, but still requires large run times for fine meshes. A programming model with CUDA and MPI that utilizes GPUs on multiple nodes can significantly reduce the wall clock time.