@Aurum420 Currently, the thought is the same work units will be run on CPU and GPU, no difference in parameters or expected results. However, it'll act as a separate application version on the system. They will basically run the same workunits and help increase the throughput of a single batch.

I think what you're asking is if a client does a request for work, how will it try to keep the GPU fed. When the client does a scheduler request, similar to beta, it'll try to send GPU work if there is enough room in the queue on the client for it. The feeder on our end will have say 500 work units ready to send with no application version defined until it is claimed by the client.

Let me know if this helps. Trying to understand what is causing concern. Happy to try to explain to the best of my knowledge :)

Thanks,
-Uplinger

I am looking forward very much to use some of my GPUs for WCG crunching!
Hope it will come soon.

We saw that ADADELTA was significantly better than Solis-Wets for molecules (ligands) with many rotatable bonds. However, for smaller molecules, Solis-Wets is actually better. We should edit the GitHub page to accurately describe the performance of the local search algorithms.

In autodock, search algorithms try to "guess" the best binding pose of a molecule in an iterative fashion, where an iteration is a call to the scoring function to evaluate a new binding pose. Some algorithms are better than others at finding good binding poses in fewer iterations.

A curiosity: ADADELTA was developed by the machine learning community for training neural networks. https://arxiv.org/abs/1212.5701

Hello :)
Yes, it's almost certain that ADADELTA will be used for the larger molecules, and Solis-Wets for the smaller ones.

I am a beginner in the community, and I would like to recruit new members for my team. I usually have 4 devices constantly working. https://join.worldcommunitygrid.org?recruiterId=898433&teamId=3QR87KXZC2