Flying Fox wrote:
Neglect in terms of writing proper SMP code (that can scale up with >4 cores) so better "science" can be done through CPUs. For folding, it is not entirely about the FLOPS, but the amount of "science" that is done. Points were supposed to be awarded that way. Given the current discrepancies of points generated per unit of GPU vs CPU, a lot of SMP-only folders are getting discouraged, especially the BigADV guys who are being forced back to "puny" SMP projects.
Not to defend everything they've done... but the CPU work they're doing these days isn't highly parallel to begin with. All the projects types that benefit the most from parallel processing were migrated over to the GPU.
This was the problem with Bigadv folding, the SMP core & projects didn't lend themselves to multiprocessing well. A 4P system in reality folded around 3.5x instead of 4x of a single CPU. That was the specific reason they started the BigADV program, was to add enough bonus points to offset the performance loss. The entire thing got out of hand and they ended it badly... but I do agree it needed to be ended as it was simply harming Folding overall at that point.
They should do what the coin mining people do and develop custom circuitry for this stuff that are a hundred times more efficient. Then all the Mother Theresas of the world could buy a couple and stick them wherever they please (wall socket or spare USB port) and "help humanity" without burning the goddamn house down.
That isn't physically possible. ASIC hardware requires what is being computed to not change... in effect, they would have to design a new ASIC every single time they added new projects which is almost weekly.
If you want to understand why it isn't possible, go read some Ars Technica posts on ASICs and coin mining.