It's still a pretty new app, hopefully they'll continue to improve it.
WCG Beta Testing GPU Work
- Thread starter doneske
- Start date
They are only giving out 2,000 work units every 30 minutes.
Is that "temporary" or permanent? That's not really that many.
It's the until further notice rate.
I've been following the official forum for a while now and they haven't given any detail as to why. Virtually nothing official has come out since the stress test. Can they, IBM or Scripps Research, not handle more work than is already going out? I wonder why there are even CPU work units still as I was given the impression the tasks for GPUs are the same kind of work. Are the CPU tasks actually ones that can't run on the GPU? Others have asked these as well. Hopefully we'll get some information soon. We are past due for an update via Research Call Notes on how the project is going in general.
I've been following the official forum for a while now and they haven't given any detail as to why. Virtually nothing official has come out since the stress test. Can they, IBM or Scripps Research, not handle more work than is already going out? I wonder why there are even CPU work units still as I was given the impression the tasks for GPUs are the same kind of work. Are the CPU tasks actually ones that can't run on the GPU? Others have asked these as well. Hopefully we'll get some information soon. We are past due for an update via Research Call Notes on how the project is going in general.
I observed today that I was getting CPU work. I don't see any reason to commit CPU resources now that there's a GPU app so I disabled that.
Here they said this: "Currently, only about 20 percent of World Community Grid power comes from devices with GPU, so participation from every willing CPU-enabled computer, Android device, and Raspberry Pi remains crucial" (https://www.worldcommunitygrid.org/about_us/viewNewsArticle.do?articleId=693).
A five-year-old, low-end GTX 1060 still does 30 times as much work as an enthusiast-grade AMD 3900X consuming about the same electrical power. There is many times more GPU processing power available right now than CPU processing power. It feels like a major waste to run CPU tasks for a project that theoretically gets the exact same work done with GPUs. There are plenty of good projects that have work that can't be run efficiently on GPUs at all.
Last edited:
I am still running some Android tasks as there aren't many projects that will work on my devices. I don't intend to run any more standard CPU work though, unless there's some kind of competition.
As you know I have been scaling back a lot of my hardware. I donated 4 servers to the iThena project and getting those things through EU customs was a nightmare. I would like to donate something to TN-Grid but getting the equipment to Europe is not something I want to tackle again. I have two more HP 1U servers I'm going to sell at some point and then I will be left with the 3 home built systems and the big Dell server. I think I've settled on putting GPUs in the 3 home built systems and running only GPU work on those and CPU only work on the Dell. Unfortunately, with the scarcity of GPUs and the associated high prices, it looks like I'm going to be CPU only for awhile longer (other than the Radeon Pro W5500 I bought to become familiar with setting one up on Linux).
I expected to have multiple upgrades done by now but even the cost of standard components is up. Nvidia GPUs are impossible to get locally. AMD cards are available at exorbitant prices, e.g. a 6900 XT is over 2 grand. I'll pass.
Hopefully things have settled down by next spring or summer. In the meantime I'll keep plugging away with what I have. I've slowed down a lot anyway as it's gotten warm and I had to put the air on sooner than I hoped.
The May Update further leads me to believe CPU work is just so people can contribute and not different work from that which runs on GPUs. Also says the rate of GPU work is 1,700 work units per 30 minutes but further down the thread they reference Uplinger says it was increased to 2,000.
"The scientists for OpenPandemics - COVID-19 provided us with approximately 30,000 batches of GPU work (equal to the amount of work done in about 10 months by CPUs), and we let these batches run until they were fully processed."
And below: "35,296 batches in the last 30 days (largely due to the stress test)".
So, approximately, 10 times faster, right?
And below: "35,296 batches in the last 30 days (largely due to the stress test)".
So, approximately, 10 times faster, right?
Those 30,000 batches for the stress test were completed in eight days. In eight days the 20% of systems with GPUs contributed as much work as all systems with CPUs could over 10 months. So 187.5 times faster...with hours of downtime, issues downloading/uploading work, and other hiccups. During the beta GPU batches were completing 334 times faster on average. So we can estimate that there is 37.5-66.8 times the processing power available to OPNG vs OPN1.
Yes, you're right! Unbelievable! Thank you!
Now I understand what you have been saying: these people who want to contribute and don't have GPUs could easily start to contribute to other CPU only projects available in WCG or elsewhere.
Uplinger et al. would be best served scrapping the current CPU app at this point and focusing on GPU work issuance and development, in my opinion. This is like using wheelbarrows when you have a fleet of dump trucks available. There's no shortage of CPU work for those without a GPU, both WCG projects and others.