UK BOINC Team Forum

The software that powers OpenPandemics - COVID-19 has been adapted to use GPU power. This will not only help the researchers to screen more molecules, but also could help them study molecules that are more complex.

https://www.worldcommunitygrid.org/abou ... =115495183

The new GPU application for OpenPandemics will be sending out 1700 work units on average every 30 minutes.

https://www.worldcommunitygrid.org/foru ... read,43311

Hi Chris,

I'd ran out of useful work on my GPUs today and had WCG as the backup project so I've managed to grab a few.

There's not many and they don't take long (less than five minutes) but hopefully that will change.

Yeah I think this is just the first Alpha phase.. they are going to release more, with harder/longer computations as time progresses.

The tasks don't seem to load the GPU much, I'm getting intermittent spikes to 55 degrees but it's mainly sitting in the low 40's. Collatz pegs it at a constant 82 degrees

I'm running 8-up on quicker CPUs, and 4-up on really slow ones (2.5GHz) via app_config, allowing a full thread per tasks; this is running on 10 and 20 series Nvidia GPUs. I am also still running the update script.

Hi all

I'd just like to mention that UBT is actually doing quite well in the current round of various WCG Challenges...

https://www.worldcommunitygrid.org/team ... 9TGC4G1RP1

This is mainly due to results turned in by Chris C and Mark but with plenty of other contributions adding to the "cause":

https://classic.boincstats.com/en/stats ... /12/0/6429

One would hope that the GPU tasks now being made available (on the OpenPandemics sub-project) will benefit the research being done (as results can be obtained more rapidly using GPUs than with CPUs).

regards and thanks to all supporting WCG tasks.
Tim

Right now there is a stress test going on at WCG on the Convid GPU application, with the result being that they are pumping out an excessive amount of tasks. The tasks are doing real research (substantially more than the cpu tasks), but once the stress test ends (probably 2 days or so) task availability will go back to the normal trickle.

https://www.worldcommunitygrid.org/foru ... read=43386

If you run 4-8 tasks per GPU (via app_config), allowing a full thread per task, as well as run the update script (Windows or Linux) then you can do some SERIOUS numbers at WCG.

To put this in context, currently I am doing around 10% of my total credit (which has taken months/years), in one day eg 10+ million.

The update scripts I am using are:

Windows:
Linux:
Please note: the updates script will become more necessary once the stress test ends and tasks become limited.

I didn't realise there were WCG challenges currently running, it's good that UBT is doing well

I'm also running 8 up on my 20 series GPUs, my 16 series peaked at 3 per GPU (*4 worked but didn't give enough gain even with a 4GHz CPU, maybe it was the lower amount of GPU memory?)

I need to swap a few GPUs around to get my 10 series into hosts with enough cores. Luckily it's a long weekend

I was experimenting with WCG while Minecraft was offline but back on there now for another day to get my 100 million then I'll give WCG a proper go, hope the tasks last until then!

No update scripts needed at the moment but I'll make sure to add them after the stress test finishes, thanks Chris.
Adjust <gpu_usage> to give the required number of tasks per GPU (1/n)

Woodles wrote: ↑Sat May 01, 2021 11:45 am I didn't realise there were WCG challenges currently running, it's good that UBT is doing well

Hi Mark

There are usually WCG Challenges running most months...and I try to log in to my WCG account to ensure UBT are taking part...as each Challenge must be "joined" separately.

The Challenges can be created by any team and can involve any specific sub-project(s) and each teams ranking position can be based on a number of different factors, such as the number of WCG points earned or the number of tasks completed...

UBT tends to do quite well, but this varies depending on which team created the Challenge to begin with.

regards
Tim

Its slightly awkward atm calculating GPU optimisations as there are 2 different sets of GPU tasks: On my hosts I noticed quick 2+ min ish tasks, as well are more recent tasks that are taking much longer at around 10+ mins.

But something to be aware of, very occasionally tasks seem to just fail but carrying on running (but not actually doing anything); longest I have found so far is over 3 hours. Personally I just abort anything that I see over an hour (as admin mentioned that some tasks could potentially run to an hour with the right conditions).

I am running 8-up on all my GPUs, apart from the ones in the slow V3 Xeon, and on the Radeon VII - for some reason (not todo with the CPU) this GPU just could not handle 8-up; tasks went from taking 10 mins to over an hour, so I switched it back to 4-up and runtimes went back to normal.