One way to speed up calculations is to perform them not on the central processor (CPU) but on the video card. Powerful graphics cards have computing units (GPU) that can do the job of the processor. They are not as powerful as the cores of the processor, but there may be a lot, now thousands of them.
Unfortunately, the molecular-mechanical tasks are almost impossible to scale thousands of times (this is true both for video cards and for clusters of computers). The more the model, the better it is scaled. On the other hand tasks containing only tens or hundreds of atoms are scaled so bad that they can even slow down on the video card. In this case it is necessary to return to the processors. Typical tasks include approximately 1,000 atoms. This is a sad fact. Such tasks already are hard enough for the CPU, but at the same time, small for the GPU. The only solution is to consider several of these tasks in parallel, for example, by running a number of separate applications.
We performed calculations on the GPU in the program Abalone. It allows to work on the GPU with models containing up to 10,000 atoms 1) with acceleration factor of 10-40. When working with 1000 atoms the acceleration factor of 2-4 is usually.