F@H and World Community Grid

[sergd]

I am considering switching from World Community Grid to F@H, primarily because I can use my GPU with F@H in addition to CPU. But in general my concern is squeezing the most science benefit from the hardware that I have (Q6600 and 8400GS on Vista 64). So far I had limited success with GPUs and F@H. I am trying to use a fanless GPU to minimize noise, so I tried Radeon 4350 and now trying NVidia 8400GS. Radeon would start resulting in Folding errors in about 15 minutes (sounds like a known issue from what I saw on the internet) and then system reboots. NVidia seemed to be more stable, although the system rebooted once overnight as well and most importanly causing 2D drag issues under load. SMP client (ran without a GPU load) resulted in a blue screen as well after a while. The system was in use running WCGrid at 90% CPU load for more than a year, going 10-20 days without a need to reboot. I am willing to troubleshoot furhter, but I want to make sure there are good reasons for switching. More specifically:

* Besides having GPU client what are other scientific / society benefits of running F@H comparing to WCGrid?
* Is SMP folding code more optimized than Boinc / WCGrid? I have noticed that Boinc keeps all cores loaded up to the percentage I specify, where SMP F@H client keeps the CPU loaded only up to 30% with no limits imposed on my side.

[kelliegang]

I only had a quick glance at the World Community Grid.... [by the way... the PG would benefit from having a look at the layout of the website and the ease of use]

To me as a layperson it would seem that the World Community Grid is about specific goals... whereas the work of Folding@Home is about more general understanding of the underpinning science to a plethora of problems [anything related to protein folding or protein buildup]


As for your card ... I believe 8400 is a very low end card in terms of folding@home capability [if it's capable at all?] The Q6600 should be more than capable of SMP folding though? I'm sure some of those technically minded people will chime in shortly

[John Naylor]

The 8400 is one of the slowest FAH-capable cards, so you're probably better off just running CPU clients if the GPU client is crashing your PC. As for your questions:

1) I suppose that depends on what you want to study. WCG has sub-projects that can target different areas, and are therefore quite often researching a narrow area (say for instance, just AIDS). Folding@home researches protein (mis)folding, which has implications for the fighting of most diseases, and is therefore of great importance for developing vaccines for many of these diseases (although as of late there has also been some more targeted research by this project).

2) The current windows SMP core (the a1 core) is not particularly well optimised. Ideally it should use 100% of your processor, but for various reasons it does not. Its replacement (the a2 core) has already been released for Linux and Mac OS X, and this is much more efficient, using almost the entire processor. However due to problems getting the underlying scientific code (Gromacs 4) to compile on windows, development of the new core for windows has stalled. If you want to use 100% of your processor on windows the best way is to install multiple single-processor clients (4 in your case).

[sergd]

Regarding #2, I think I have figured why my CPU was not loaded at 100%. I didn't realize I needed to specify "-smp" in the command line parameters. Once I specified, now it's fully loaded and utilized.

Thanks for everybody who commented!

[MtM]

* Besides having GPU client what are other scientific / society benefits of running F@H comparing to WCGrid?
* Is SMP folding code more optimized than Boinc / WCGrid? I have noticed that Boinc keeps all cores loaded up to the percentage I specify, where SMP F@H client keeps the CPU loaded only up to 30% with no limits imposed on my side.

Both projects do good, let me say that first. But there are benefits f@h offers which boinc does not.

1. F@h offers smp clients, I don't know of any boinc projects which currently supports those ( but they may be around now finally the technical support was mentioned some times already in the past, just no projects used them ).
2. F@h's HPC clients use a diffrent approach then most boinc projects, a single wu is only folded once ( if succesfull, if not it get's send out to others as well ). Most boinc projects have to run the same work unit more then once. Furthermore f@h is much more serial, therefore the bonus with the hpc versions in ppd, every wu which is turned in quick leads to a new gen ready to fold. This approach is allot more efficient then the one used in the past with the uniprocessor clients and as far as I know with most other distributed computing projects.

What's also important but I'm not sure how boinc handles this, all research from f@h is available to the public/fellow researchers which means there isn't a single commercial entity involved with it which aim would be to make profit with the results from our contributions.

The main drawbacks well you mentioned it yourself, the sometimes problematic/confusing/unfriendly behaviour of the hpc clients. Though, it's important to mention that for the most part they work reliable imo, and allot of issues are probably more user error. Or, not error, but unfamiliarity such as for instance those trying to gpu fold without reading a guide/faq, who use uncertified/unapproved drivers, who install the smp client without reading a guide/faq ( and forget the -smp flag ). Then, the windows smp situation is more confusing with two possible dependancies, mpich2 or deinompi, while the linux/osx clients are I'm told allot less error prone/bug ridden so it might even be not the clients fault but more related to the os itself.

I think the benefits outweigh the drawbacks, and I'm happy to hear you decided to give the project another try. I hope it's more succesfull then the last one, and if you do run in any issues this forum is always happy to assist in troubleshooting.