What Do You Want from the Points Revamp?
We touched on this a wee bit in my FAH@GPU thread, but I think it might be interewsting to go into a little greater depth. They Stanford boys have been hinting at a revamp of the points system for quite some time on account of the outdatedness of the single core benchmark machine and the rise of high performance clients on GPUs and the PS3.
That makes me think about how they're going to structure the system to
- Be fair to contributors, however they choose to perceive that
- Be respectful of the existent points of current contributors, e.g. don't suddenly make the benchmark score 10k PPD, thus making past contributors think their effort was chopped liver
- Add incentives for people to use machines and clients that give the best scientific performance, not just points
That last one is particularly relevant to me. I had a go with the GPU client which is about 10x as productive as a standard client. I usually run the Linux MSP client on the machine. We'll say that that gives 4x the performance of a standard client to be generous. The GPU WUs give me less PPD than the SMP WUs. If the GPU WUs scored better they'd be getting more than twice the productivity out of me in terms of scientific output.
How would you enterprising chaps address the problem?