Hi folks. I thought I would put this up for anybody interested. I performed evaluation of a partition function related to molecular magnets using the 5850 under OpenCL (Stream SDK 2.0), and compared the results to running the same code on the CPU, and a similar serial implementation using the GNU Scientific Library (GSL). I think the results speak for themselves - the 5850 wins hands down! Hope somebody finds this useful. I will eventually get around to testing this on other systems and devices, and update it to suit the newever releases of the SDK.
P.S. The above document has already been submitted, so if you spot any grievous errors there won't be a lot I can do to rectify! Feel free to let me know though.