Today we’ve had the first meeting of the group with the goal of accelerating radiative transfer calculations around black holes with graphical processing units (GPUs). The idea is that by porting the code to exploit the massive parallelism of GPUs, we will see a dramatic speedup (hopefully ~10-100x). We have an enthusiastic group of students!
We decided to proceed with NVIDIA CUDA rather than OpenCL due to the better documentation and more astrophysical examples available in CUDA to guide us.
We will try to have a beta version of the GPU-accelerated code by December, and a release candidate on May 2018. The main challenges–as usual with GPGPU–is writing the appropriate kernels and GPU memory management in order to reduce host-device data transfers as much as possible.
Stay tuned for exciting news on astrophysical high-performance computing by our group!