@joelkuiper: I put Bayadera on github (http://github.com/uncomplicate/bayadera) so you (or anyone interested in MCMC on GPU) might preview how it takes shape. Be warned: this is a very, very, very early preview. It still has support only for univariate distributions. Only a handful of distributions are built in. Everything can (and will) change daily. It is not documented AT ALL (if we do not count tests), since there are lots of changes. On the other hand, what is there works well, there are tests and code examples from the best books available (Doing Bayesian Data Analysis), and it is relatively easy to add your own distributions to the engine (once you understand how the existing ones are created). The next big milestone is the multivariate MCMC engine. After that, everything else can start to stabilize its shape.