I spent some time over New Years looking at this.  Here's my take:  There's something cool in having a CPU-profitable coin to mine.  If everyone goes GPU, it'll change that.  But at the same time, it does seem like there's at least one private GPU implementation for PTS, which is great for the person who has it, but contributes to their ability to mount a 51% attack on the network.  If you actually care about the coin instead of simply mining, that's a bad thing.  If you just mine, you don't care. :-)  Since you can't put the cat back in the bag, the other option is to let everyone GPU mine.

Which brings us to:

I can write a GPU miner for protoshares.  It'll take some very solid effort on top of the sketch I've already made of how to do it.  I don't really have that time, but I'm willing to do it for a bounty.  I would propose a first-cut CUDA version integrated with some existing miner, in a form that someone like Christian (of CudaMiner) could take and add his auto-tuning magic sauce to to really eke out better performance over time.  Whomever writes the polished version can get compensated for their time by asking for tips on startup or by adding an auto-tipping mechanism similar to the one in yam.  The algorithms will be fairly straightforward to convert to OpenCL code for use on AMD cards, and I'll do a full open source / Apache 2 licensed release of the source code along with documentation.

Are people interested enough to fund this at a level that justifies the time investment?

(To back up my claim a bit, recall that I'm the person who wrote the new-improved Kepler based scrypt mining code for NVidia GPUs, and I seem to have a day job as a Ph.D. computer scientist.  When I say I can write the miner, I'm saying it with absolute certainty.)

  -Dave