Due to its great computational abilities, our GPU2 client has had a great scientific impact so far. In Paper #72 (also see the movie), the GPU clients play a star role in allowing Folding@home to push to unprecedented levels, simulating atomic-level protein folding on the millisecond timescale — several orders of magnitude longer than any previous atomistic folding simulation.
GPU3 brings several key new features to Folding@home. In particular, GPU3 allows for greatly enhanced science: including more accurate models, new science can be done, 2x faster execution of the science, more stable simulations, OpenCL support for run time science optimizations, and greater flexibility for adding new scientific capability. This is accomplished through the use of the https://simtk.org/home/openmm/OpenMM GPU library (which originally came from FAH GPU code, but has been significantly enhanced by Simbios staff).
GPU3 also lays down the foundation for future incorporation of OpenMM’s support of OpenCL, which will also bring some very important new scientific features, especially in terms of on-the-fly runtime optimizations of the scientific code. However, at the moment, OpenCL is not supported in the GPU3 NVIDIA client.
Since GPU3 appeared for NVIDIA first (and for a while on NVIDIA only), we have used an NVIDIA GPU for benchmarking. See the Points FAQ for more details.