GPUGrid Work Availability

[Nick Name]

I might make this a sticky so it's easier to keep track of what's happening with this project since we've been one of the top teams pretty much since its start. Over the last year work has been kind of erratic especially in the last few months. The project has had some turnover in developers and came out with a new app that took awhile to get most of the bugs out. Work has been in short supply for quite awhile. As of now there's plenty of work, the server shows thousands of available tasks for the first time I can remember in quite awhile!

If you want to participate, review the project requirements here.

FAQ - Acemd3 application and driver versions

www.gpugrid.net

TLDR:

1. This is a Windows and Linux Nvidia-only project.
2. You must select the ACEMD3 app in your project preferences.
3. Windows, CUDA80 Minimum Driver r367.48 or higher
   Linux, CUDA92 Minimum Driver r396.26 or higher
   Linux, CUDA100 Minimum Driver r410.48 or higher
   Windows, CUDA101 Minimum Driver r418.39 or higher
4. This project can push GPUs very hard compared to other projects. If you card is overclocked and work is failing, back off. Just because it's stable on other projects or games doesn't mean it will be stable here. Also make sure the card isn't running too hot, you might need to set fans speeds higher than normal. I think Nvidia cards, at least from Pascal up (10xx series) start to throttle at 72 degrees so I try to keep mine below that.
5. There is a bug in the app that affects systems with different GPUs. Using a system with a 2080 and a 1080 as an example, if you need to restart BOINC or your computer, if the app was running on the 2080 and restarts on the 1080 it will fail. There are some ways around this. I feel the best solution is multiple clients. If you have a system that has an older Nvidia card, like anything older than Maxwell (9xx series), a project exclusion (<exclude_gpu> in your cc_config) might be a better option.

The current work announcement is here.

My machines were built primarily to crunch this project so I'm excited about this.

 

[Nick Name]

This batch ran out in just a few hours. Hopefully that was a test run for a bigger batch.

 

[Nick Name]

Work is still spotty at best. Hopefully these test jobs pave the way for larger tasks and consistent work soon.

 

[Nick Name]

I just realized I hadn't updated this thread. There is now PLENTY of work...except for the current outage.

I came home to find every client dry and trying to upload finished work. The site is down, I saw a message on Twitter they (I expect Toni) were rebooting the server. Hopefully it's not a major hardware problem.

 

[Nick Name]

Work is still readily available and should be for at least several weeks, if not months.

 

[Nick Name]

Work is still plentiful. I've put some resources into our new Folding@Home team, but with the recent surge of interest there due to the coronavirus work is extremely difficult to get. I've set exclusions in cc_config and set GPUGrid to a share of zero to deal with the FAH downtime. Excluded apps for me are as follows:

    <exclusive_app>FahCore_21.exe</exclusive_app>
    <exclusive_app>FahCore_22.exe</exclusive_app>

How this works: When the Folding@Home apps are running, BOINC suspends GPUGrid work, and resumes it when Folding work isn't running. This works very well if you want to keep your GPU crunching / folding as much as possible. You could use this for any BOINC project, at first I was using PrimeGrid (sieve app) as the backup since the work runs quickly. Now I'm using GPUGrid as the FAH outages have gotten longer and longer.

 

[Nick Name]

There's still plenty of work available, but the admin Toni recently posted saying the number of hosts attached has doubled in the last month. I'm seeing work slow to upload and download at times, and the forum is very sluggish. There's been a longstanding network issue that is thought to be in the university or possibly an ISP backbone provider that has been causing problems like this for a long time, and the increasing load is making it worse.

 

[Nick Name]

There's still plenty of work available, but the admin Toni recently posted saying the number of hosts attached has doubled in the last month. I'm seeing work slow to upload and download at times, and the forum is very sluggish. There's been a longstanding network issue that is thought to be in the university or possibly an ISP backbone provider that has been causing problems like this for a long time, and the increasing load is making it worse.

 

[Nick Name]

Work is still plentiful but the server keeps running out of storage. This has been happening about once a week for a month or so. If you can't get or return work that's probably why.

 

[Nick Name]

The last batch just completed so there won't be any work for awhile.

Summary of the MDAD effort

www.gpugrid.net

So, the huge MDAD batch that we sent at the end of January just finished crunching. That was by far the largest continuous load we placed on Gpugrid, so thanks everybody!

I will be sending more continuation workunits, but it will take some time to arrange the details. For the time being, take it as a deserved break

The objective of the batch was a large-scale exploration of the protein landscape. Its accumulated 25 ms of sampling will be useful for biomedical protein modeling at large (we'll provide more details in due time).

Crunching the whole set was a technical challenge per se. Volunteered power has grown a lot during the project. This may have been motivated by (1) the continuous availability of so many workunits; (2) shutting down SETI server; (3) the lockdown; (4) the notoriety of distributed computing for fighting COVID-19. This placed additional strain on the server, but in the end the mammoth task was completed - and we want more

T

 

[Nick Name]

Work is available again. The work server has also been converted to HTTPS, so you'll get a message about using an outdated URL. The easiest way to fix this is run out any work you have, detach from the project then attach using the new https address. Save any app_config or project specific files you want to keep that are in the project folder since it will get deleted when you detach.

 

[Nick Name]

There's been no new work for a few weeks now. I haven't seen any news from the admin on what they're working on or when new work will be coming.

 

[Nick Name]

Nothing new to report, and there's been no word from the admin for quite awhile. I don't think the project is dead, there's quite a bit going on with the new Nivdia GPUs and the virus and I expect that's slowed things down like it has everything else.

 

[Nick Name]

There's still been no update from an admin, but there is some new work coming through. These tasks are extremely large, taking around ten hours plus on a 2080 Ti. If you try them keep in mind GPUGrid has a short - 5 days - deadline as new tasks are made based on results from tasks that are returned. This work may be Linux-only, I haven't gotten any work on my Windoze machine.

 

[johnnymc]

There's still been no update from an admin, but there is some new work coming through. These tasks are extremely large, taking around ten hours plus on a 2080 Ti. If you try them keep in mind GPUGrid has a short - 5 days - deadline as new tasks are made based on results from tasks that are returned. This work may be Linux-only, I haven't gotten any work on my Windoze machine.

I am crunching these on two windows boxes.

My 2080 takes 15 hours and my 1080 takes 23 hours.

 

[Nick Name]

These run times remind me of the old days when I first started this project. There seems to be plenty of work right now if your system can handle these jobs.

 

[Nick Name]

Work is out again, that batch didn't last very long in spite of the task size. It will be interesting to see how it goes when the 30xx series cards are more readily available!

 

[Nick Name]

Work is sparse, I was dry for a few days then a couple tasks yesterday.

 

[Nick Name]

Work is still sporadic, maybe a couple a week.

 

[Nick Name]

I had 2 yesterday.