Free cpu time on server for remote folding

Case number:671076-1998865
Opened by:Bruno Kestemont
Opened on:Sunday, December 7, 2014 - 12:31
Last modified:Wednesday, April 13, 2022 - 18:30

Imagine a Foldit genius with a slow computer.
What if he/she could get access to a remote powerful computer just a few hours a week, only to improve it's solution with scripts?

I know it's contrary to the starting idea of Boing (using cooperative decentralized cpu units). But nowadays, the central computers are less expensive, and the personal computers are evolving to slow machines with high communication potential, with a strategy using remote powerful central units and stocking devices (clouds). Like iPods and other tablets. Our computers are turning again to a kind of terminal more than supercomputers.

If the humans are still better in folding in 3-D than algorithms, then giving access to a powerful machine to best players could help find solutions.

A limit (and an incentive) could be the following:
-limit to a few hours a week: player has to focus it's use of this facility to the most promising script at the right moment of the game;
-limit to the "hall of frame" best soloist players, e.g. players with >30000 hall pts.

This is to be sure that only very experimented players would use this facility, and only on purpose, with parsimony.

Yes I know, it's seems not to be fair for the other players (beginners etc). But I think one of the game characteristics is the fame, the possibility to go ever "higher" in reputation, as a recompense linked to the merit and hability. Exactly the same rule is applied for beginners being obliged to demonstrate minimal learning before to go the the "real science" puzzles.

We could compensate this by giving a kind of "beginner's bonus" to the other players, or contrarily giving a penalty in points for each second using the server CPU.

Other ideas are welcome (is there something like a "free cpu cloud" in google or others?)

(Sun, 12/07/2014 - 12:31  |  3 comments)

Joined: 05/19/2009
Groups: Contenders

The underlying idea is that the client has become too slow.

Fixing this will allow more people to participate. One way to fix the speed issue is to optimize the algorithm and to code portions of slow code in OpenCL, as I have been promoting (and offering) for several years now.

A remote powerful computer (single core) for a few hours per week gives you a certain number of GigaFLOPS, probably less than what you would have when leaving a weaker PC running all week.
OpenCL is the way to go, combined with cheap AMD GPU hardware. Anyone should be able to affort a 180 euro videocard that can boost both video performance and GPGPU performance.

If people cannot spare this for a videocard I think they have more urgent problems than to waste their time folding proteins.

Joined: 06/06/2013
Groups: Gargleblasters

Bletchley Park is only partly correct. He is correct that the client is too slow. Puzzle 1024 has been jamming up CPU unless you shut it down and start a new client frequently, and it is still verrrrrrry sloooooow.
But Bruno is also correct. We have some intermediate and top notch players on my team (Gargleblasters) that are severely restricted to only one or two clients. While one is a superb hand folder, the rest of us struggle. I know other players run 7 or 8 clients. Also, larger teams can divide and conquer.
If our goal is to get the best science, I think perhaps some safely segregated CPU would help our better folders compete. My team looked at CASP as a test of who had the most machine rather than who was the best folder. Most of us had to pick and choose some puzzles, as we could not do them all. While I would classify myself as an intermediate player, I would also say I am very frustrated
We can't use evo status as the cut off, as AD has blocked other teams out of the top spots pretty consistently (I'm not complaining -- it makes them a good competitive team) But perhaps we can use solo rank as a proxy for players with enough mileage to merit some CPU time.

Joined: 09/24/2012
Groups: Go Science
Status: Open » Closed



Developed by: UW Center for Game Science, UW Institute for Protein Design, Northeastern University, Vanderbilt University Meiler Lab, UC Davis
Supported by: DARPA, NSF, NIH, HHMI, Amazon, Microsoft, Adobe, Boehringer Ingelheim, RosettaCommons