pedwards wrote:Interesting point regarding the Core_16 vs. 17 speed... assuming that closely mirrors to v6 vs. v7 clients?
There is nothing in V6 vs. V7 to mirror. Each Project requires a specific FahCore. Each FahCore supports certain hardware and certain drivers. The goal is to design a FahCore that will work with all possible hardware and drivers, but that's a challenge every time a new line of hardware is developed.
At one time, ATI was supported on Brook drivers but begniing with the HD5xxx series, they decided to support OpenCL and soon dropped support for Brook which made the HD4xxx and earlier unsupportable. FAH developed FahCore_16 which was written for the then-current version of OpenCL and dropped support for the pre-HD5xxx hardware. AMD has continued to improve their support for OpenCL. FahCore_17 is a new FahCore that has been written to be more efficient by taking advantage of newer features in the newer version of OpenCL and available on newer hardware.
NVidia has a similar story with a different sequence of events. NV created a proprietary language "CUDA" while providing a more limited OpenCL capability (I'd add "reluctantly" but that's my personal opinion, not necessarily supported by the facts). FahCore_17 isn't working as well with this earlier version of OpenCL but it's up to NV to decide if they want to improve their support for OpenCL or not. If the only way to get universal support of NV hardware is to support CUDA, Stanford may do that sometime soon, (to which I'd again add the opinionated word "reluctantly") since it means more software to develop/support and more infrastructure to provide that support. There's no firm plan for that.
Ideally, a single FahCore could provide OpenCL folding capabilities on all GPUs and on all Operating Systems, but the OpenCL drivers on MacOS are not up to the task and the Intel drivers which are supposed to provide OpenCL support for Windows and Linux are also sub-standard at the present time. I dream of a day when NV/AMD/Intel GPUs can fold on OS-X, Linux, and Windows using a single FahCore.... but I'm dreaming.
If the GPU manufacturer can't project sufficient increases in sales by investing in OpenCL on a specific OS to make it worth their while, they're not likely to do it.
When possible, V6 and V7 use the same FahCores to fold the same Projects. In some cases, features needed by newer FahCores were added to V7 to support them so V6 is more limited. Stanford is no longer upgrading V6 but if somebody wants to do so, that's up to them. The third party wrapper was added for WINE which retrofitted V6 with limited GPU support.