WCG BETA work for GPU's

"We are using the OpenCL version (1.2) and a binary is available for NVIDIA, ATI, and Intel GPUs on Windows 64 bit, Linux 64 bit, and macOS 64 bit."
 
If you are wanting to run these on your Intel iGPU, make sure to go into the profile settings on the WCG site and change it to "yes" because only AMD and nVidia appear to be turned on by default there.
 
It looks to me that the gpu project for open pandemics is making good progress with their regular updates and regular flow of beta tasks. As with new projects, there is more preparatory work to be done. https://www.worldcommunitygrid.org/forums/wcg/viewthread_thread,43290_offset,290#654270

Hopefully we will see the long awaited gpu project for WCG very soon.

Sorry about the hiccup in work. I had to run to the store to grab something and took longer than expected for making sure we had plenty of work loaded up. I would imagine at the current pace we will be out of new beta work in about an hour.

This test is putting some pressure on the system which is nice to see how it handles the validation/assimliation of the results that are coming back in quickly.

We are having some virtual high fives here in communications with the researchers. They have some additional checks they want to look at, but currently we are getting really good results and things look awesome.

With this latest beta batch, I have changed the assimilator to be production ready with how it packages results for the researchers.

So next steps in my mind are this:

1. Eat pizza for lunch (check, yes i really had some pizza, nanoprobe knows me too well)
2. Send additional results back to researchers for validation (tomorrow)
3. Await final thumbs up from researchers (Hopefully we'll have this by Friday)
4. Build batches and upload 7.28 to opng (Tomorrow)
5. Perform other final checklists (ongoing)
6. Have go/no go conversation with the researchers (TBD)

Thanks,
-Uplinger
 
I was getting some sporadically to run on my RX570 a while back. It chewed through them pretty fast although they weren't pushing the GPU very hard. The GPU spent quite a bit of time idle and not crunching as shown by the large difference in processing time and actual work time. I was also able to see when the GPU would stop working. Some of the times it did this the CPU usage would spike for the GPU process as if it needed to process something else before going back to work on the work unit. This was under Manjaro Linux.

I haven't run any in a while since I made a motherboard and RAM change in my main system in prep for a new CPU. This caused some issues and some reinstalling which caused issues with some software including WCG. I haven't bothered to figure out what the issue is and am waiting on my new CPU to get here next week before I do full reinstalls of both operating systems and proper setup of all software. At which point I will then be doing a full reinstall of my server with the CPU motherboard and RAM I had in my main system. It will be nice to have much newer hardware in my server for the very long overdue upgrade (especially since I'll be able to use hard drives larger than 2TB) but I will miss that Q6600. It's been a tank for the 12+ years I've been running it.

On the plus side I'll have some extra processing power in my main machine with the 5800x and a lot more processing power in the server with the 2600x being moved over to it. It's not summer temps yet so I'll be able to run both for a little while.
 
Today's update on gpu processing on WCG OP project. The average speedup was 334X faster than cpu. Results were based on beta gpu tasks.

Now, let's see when WCG is going to officially release the gpu tasks. Hopefully not a long wait....Fingers crossed. ;)


1617746597644.png
 
FYI this has moved beyond beta testing to small batches and now a "stress test" that started today. From an email i got:


Dear OpenPandemics - COVID-19 GPU Supporter,

World Community Grid has an opportunity to do some highly important system testing, and we need your help!

Today, we're starting a stress test of the World Community Grid infrastructure with the help of the OpenPandemics - COVID-19 research team in the Forli Lab. The purpose of this test is to see exactly what our entire system can do when we are moving large amounts of work in and out.

For the purposes of this stress test, we've been given 30,000 batches of work units to run on the GPU version of OpenPandemics - COVID-19. These are real work units that will provide data for the project.


Since about a few hours ago I am getting steady GPU work which is a first and means a ton of points. Get them while they are hot people. No idea how many units are in a "Batch" but there seems to be a lot of work right now.
 
Back
Top