F@H vs BOINC & Rosetta@home ; using GPU

[vertigo]

Ah, ok. Well, it seems it's finally using the GPU now, so I removed the CPU slot altogether anyway.

[Joe_H]

I meant distributed computing groups, i.e. might people be doing calculations for F@H that other people are doing for BOINC or Rosetta, therefore being wasteful. IOW, I was trying to figure out if both groups are doing the same work with different software, but AFAICT based on what I've read, including one of your links, it seems not. Apparently F@H does a much finer, step-by-step analysis, whereas Rosetta simply makes a bunch of assumptions to evaluate things on a larger scale. So I guess it just depends on which approach is "better," if either is, since it's likely they each have their pros and cons and each contribute in their own way.

As you have figured, they are different approaches. Rosetta gets a final "folded" configuration, F@h is looking at the steps the proteins probably go through, and can show intermediate confirmations that may open up suitable receptor sites for drugs and how they might bind to the proteins. F@h can use results showing folded or unfolded states as starting points whether they come from Rosetta or other modeling going on.

I do know that the groups in the F@h Consortium are sharing information on what each is looking at in relation to COVID-19 so they don't duplicate effort. In addition dome of the larger scale modeling is identifying areas that are being used by someone else to investigate specific details such as possible receptor site.

From what I understand they are also in contact with other researchers over what different groups are working on, again to avoid duplication of effort where possible.

It's set to choose the number of threads automatically, so I assume it's using what's most efficient.

That is more a setup from the installer to get a usable folding configuration, especially for first time installers. It is not necessarily the most efficient. Some of the default settings are a bit of a legacy from a few years ago when few machines outside of the high end had more than 8 cores. Since the client was last updated 2 years ago, that has change somewhat.

There is code in this latest version of the client to work with the servers to still provide CPU work at a lower CPU thread count if none is available for the count requested. This works most of the time, there was a recent topic from someone who was mildly surprised to get CPU:12 WUs on his 30 core system,

[Neil-B]

That was me ... and thought it was rather cool that it happened tbh - it was better than it sitting idle imho

[bruce]

do know that the groups in the F@h Consortium are sharing information on what each is looking at in relation to COVID-19 so they don't duplicate effort. In addition dome of the larger scale modeling is identifying areas that are being used by someone else to investigate specific details such as possible receptor site.

From what I understand they are also in contact with other researchers over what different groups are working on, again to avoid duplication of effort where possible.

I have a lot of faith in scientists. They do form partnerships when it's in the best interest of everybody. The do keep track of what else people are working on. FAH scientists are not about to duplicate the work that somebody else is doing ... they're too busy trying to accomplish their goals -- and one of their HUGE goals is to make all their research public ASAP.

See https://foldingathome.org/news/
Yesterday's article about screening fragments identified by another team impressed me:
"... working with an international team of researchers to computationally screen potential inhibitors of the coronavirus...

[vertigo]

Maybe I'm overestimating the scale of impact from the outdated code, but it seems, especially with such a large number of computers being used, that the project would benefit hugely from prioritizing the updating of the code to make the most efficient use out of users' hardware. Leaving it using the "legacy" code that isn't using the computer efficiently just seems like a massive waste, resulting in a huge loss of productivity. I mean, if it's something like a few percent, that's one thing, but if updating the program to make better use of hardware, whether that's more efficient use of threads or helping ensure GPUs are being used when possible, would result in 10% or more WUs being processed in a given amount of time, that would be very significant.

[bruce]

You may be right, but it's more a matter of human resource limitations of those working at FAH.

The default allocations are meant to be an initial guess that works. Those who care enough to look into the details can do better -- and should, especially if your hardware doesn't fit into the "typical" home computer when the code making the initial guess was written.+

A machine working with a poor guess is a lot more valuable that the machine not working until the owner learns enough to set it up himself.

[vertigo]

Absolutely. It's just too bad the software isn't more up-to-date and that there's not an easy to use, built-in guide to help users to even realize they *can* make it do better, as well as actually doing so. It just sucks to think about how much less work is being done because of all the default, inefficient configurations. Yes, it's better than nothing, but still... Oh well, who knows, maybe after this is all over and things have settled down a bit, it'll happen. Maybe they'll even get some funding to help after people realize the potential for situations like this.

[bruce]

No doubt.

[Joe_H]

Sometimes you have to use the tools you've got, waiting for them to be perfected can mean you don't ever get started.

In this case "new tools" were on the planning board, but COVID-19 did not and will not wait for them.

[nivedita]

Just to chime in - I've been wondering abour running BOINC and F@H at the same time. I'm sure there will be a heavy load on the system but I'm OK with that during idle time. I just don't want them choking one another for resources. It would be ideal if there were collaboration between the apps so that one would not attempt to process a WU if the other is crunching. They're not the same, but they don't need to compete.

If you want to round-robin among different projects, BOINC (alone) is actually a better fit. BOINC is a platform for multiple distributed computing projects, and you can allocate your resources so it devotes eg 1/3rd of the time to one and 2/3rd of the time to another, and it will fully utilize your computer if at least one of those projects has work to do.

At least on NVIDIA, you can actually run both BOINC and FAH on the same GPU and it works, but it slows down both of them obviously. But if you're worried about the GPU remaining idle if FAH runs out of WUs again, that might be an option.

There was a very old project to get FAH as one of the projects on the BOINC platform but it's been shelved unfortunately.
https://foldingathome.org/faqs/high-per ... -on-boinc/