Folding Forum

A couple of questions about the new FAH Core 22:

What kind of performance improvement are we looking at for Core 22 versus Core 21?
Will Core 22 support CUDA acceleration on Nvidia GPU's at release or at some point in the future or not?
When can we expect this new core to arrive?

It is the same answer as always: to discuss Beta software, join the Beta Team.
To the extent possible, all Cores are weighted to give the same points for the same work.
Yes, it either will or it will not.
As any successful software project since the beginning of time (1942?) software is delivered when finished.

Thanks

The first version of a new core will NOT support CUDA. The bug detection/new_version iteration will proceed as described above and the OpenCL version will be released (whenever it's ready). If the develpment funds have been expended, there will be no CUDA version. If there's still money in the pot, there probably will be one.

Ok thanks, that makes sense. I currently don't have the ability to be a beta tester, but I had been seeing that things were happening with the new core, so I thought I'd ask. Thanks for your quick response.

How will CUDA accelerate folding?
Can one card that's running a WU at 80%, allocate an additional 20% to a card that's running a WU at 100%?

A FAHCore which uses OpenCL has the distinct advantage of using a single FAHCore to process any WU on any supported GPU. Obviously this minimizes the coding required by FAH's developers.

CUDA is a propriety interface that's only available for NVidia.

In fact, for NVidia GPUs. the OpenCL commands are translated internally into sequences of CUDA commands which are more attuned to what the NV low-level drivers expect. By rewriting the FAHCore into CUDA, the extra overhead of that translation step is avoided. A CUDA FAHCore can do a better job of optimizing the command sequences for the hardware.

The core is in beta, please use the beta thread to talk about it : viewtopic.php?f=66&t=31445
viewtopic.php?f=2&t=10364&p=102182#p102182