At present, the only up to date GPU code in EMAN2 uses GPU-enabled Python libraries. Specifically, the new neural networking code for particle picking, heterogeneity analysis and automatic tomogram segmentation all make use of Theano which is fully GPU enabled. To make use of this, or any other GPU capabilities we manage to implement, you will need to have a current version of CUDA installed and working on your computer. More details are available on the Tomogram Segmentation Tutorial page.
EMAN2 has had more complete GPU support twice in the past once about a decade ago, and again (new implementation) about five years ago. Both of these implementations are now obsolete, and provide a cautionary tale to GPU aficionados. I can still run EMAN2 productively on a 7 year old Linux cluster, but how old a GPU can you use when you run GPU-enabled software? In any case, we may again implement some specialized programs using the GPU, and if so, we will update this page.