| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Beta Testing
Forum: Beta Test Support Forum
Thread: OpenPandemics GPU Beta Test - March 12 2021 [ Issues Thread ] |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 160
|
|
| Author |
|
|
Richard Haselgrove
Senior Cruncher United Kingdom Joined: Feb 19, 2021 Post Count: 360 Status: Offline Project Badges: 
|
Mine are in i5-4690 CPUs. They're running fine at Einstein@Home, but I was using an older driver. I've updated the drivers because they might have been the problem here, but I didn't get to them quickly enough to test the new drivers with the most recent Beta run. I'm going to try and set up in advance for the next run, and perhaps save a few tasks for offline testing - depends how much warning we get of the next run, and how long it lasts before the tasks run out. (Do we have a server status page here? I haven't found one yet.)
|
||
|
|
Grumpy Swede
Master Cruncher Svíþjóð Joined: Apr 10, 2020 Post Count: 2068 Status: Recently Active Project Badges: 
|
I haven't found any server status page here, either. And whenever the next batch is released, going by how it's been so far in the GPU Beta, all tasks will be gone after just an hour or two.
----------------------------------------To be able to test things thoroughly, we need many, many more tasks, than the batches we've seen so far. Edit: I have my doubts, that the project will even be able to churn out as many tasks as will be needed, when this goes into production. If my GTX980 crunces one task in 3-4 minutes (and the latest ones were production tasks), then more powerful GPU's will when running multiple tasks crunch them in less than that. Multiply that with all GPU's on WCG, and there's going to be one he** of a demand. [Edit 4 times, last edit by Grumpy Swede at Mar 13, 2021 9:12:54 PM] |
||
|
|
PMH_UK
Veteran Cruncher UK Joined: Apr 26, 2007 Post Count: 761 Status: Offline Project Badges: 
|
There are re-sends from time-outs on slower systems such as my old Intel GPUs that don't always complete in time.
----------------------------------------Paul.
Paul.
|
||
|
|
nanoprobe
Master Cruncher Classified Joined: Aug 29, 2008 Post Count: 2998 Status: Offline Project Badges: 
|
Last few have all been time out resends.
----------------------------------------Edit: I have my doubts, that the project will even be able to churn out as many tasks as will be needed, when this goes into production. If my GTX980 crunces one task in 3-4 minutes (and the latest ones were production tasks), then more powerful GPU's will when running multiple tasks crunch them in less than that. Multiply that with all GPU's on WCG, and there's going to be one he** of a demand. The # of CPUs here far exceeds the # of GPUs and there is plenty of work for them. I don't foresee the # of GPUs to be an issue with supply even with the faster speeds. Never happened on the HCC GPU project.
In 1969 I took an oath to defend and protect the U S Constitution against all enemies, both foreign and Domestic. There was no expiration date.
----------------------------------------  [Edit 1 times, last edit by nanoprobe at Mar 13, 2021 10:49:52 PM] |
||
|
|
geophi
Advanced Cruncher U.S. Joined: Sep 3, 2007 Post Count: 90 Status: Offline Project Badges: 
|
My Haswell iGPU in Linux received 5 of the resends today from the March 12th batch, and all completed successfully. This unlike the ones it ran on actual March 12th where they all failed with maximum time exceeded. The initial time estimates were much more realistic on the resends today.
----------------------------------------[Edit 1 times, last edit by geophi at Mar 14, 2021 4:46:55 AM] |
||
|
|
[AF>Libristes]Maeda
Cruncher Joined: Sep 1, 2011 Post Count: 43 Status: Offline Project Badges: 
|
For information: my HD4600 is part of a Haswell 4770S: https://ark.intel.com/content/www/us/en/ark/p...cache-up-to-3-90-ghz.html
|
||
|
|
goben_2003
Advanced Cruncher Joined: Jun 16, 2006 Post Count: 145 Status: Offline Project Badges: 
|
@nanoprobe I'll reset your hosts when we get more coming out. For now the beta is paused while I review more data. Thanks, -Uplinger One of my machines may need reset too, it looks like 14 tasks on that machine had the exceeded elapsed time limit error.  |
||
|
|
goben_2003
Advanced Cruncher Joined: Jun 16, 2006 Post Count: 145 Status: Offline Project Badges:
|
Worth noting is, that if you run CPU tasks at the same time as you run a iGPU task, you will slow down both the iGPU task, and the CPU tasks considerably. This was something we discovered on SETI@home, where we were running tons of different OpenCL GPU apps, and WU's. You can get away with running 1 or with luck 2 CPU tasks on a lets say 4c/8t CPU, but no more than that, or you will slow down all WU's considerably, including the one on the iGPU. That is true at least for the iGPU HD4600, if the same applies to newer Intel iGPUs, is unknown to me, since I do not have any newer CPU than the i7-4790K. Look here for some data and explanations: http://lunatics.kwsn.info/index.php/topic,1735.0.html Thank you @Grumpy_Swede for pointing this out in an earlier beta. It inspired me to run many tests and gather a bunch of data.  As someone else pointed out, those were tuned to have max performance. I ran those tests and got severe slow down of those test tasks. They did not correlate well with OPN though. While I did get slowdown running OPN on the 8 CPU threads and 1 iGPU, it was moderate, not being more than an order of magnitude slower like with the test tasks. Job sizes vary, but with a large sample set should have the average close enough. With that in mind, here is what I got on Windows 10 64 bit, I7-8650U iGPU UHD 620 for: CPU only: .73 x 8 threads = 5.84 jobs/hr GPU only: 66.3 jobs/hr CPU + GPU: 52.5 jobs/hr on the iGPU I did not record CPU during this time, however even at the CPU only rate, the combined total is less than just running the iGPU. |
||
|
|
Richard Haselgrove
Senior Cruncher United Kingdom Joined: Feb 19, 2021 Post Count: 360 Status: Offline Project Badges:
|
Intel® Core™ i7-8650U Processor
Intel says that is a mobile, 4-core, 8-thread processor with a default TDP of 15 W - so hyperthreading is in operation for the CPU tasks. I think your results need to be considered in that context. |
||
|
|
goben_2003
Advanced Cruncher Joined: Jun 16, 2006 Post Count: 145 Status: Offline Project Badges:
|
Edit: I have my doubts, that the project will even be able to churn out as many tasks as will be needed, when this goes into production. If my GTX980 crunces one task in 3-4 minutes (and the latest ones were tasks), then more powerful GPU's will when running multiple tasks crunch them in less than that. Multiply that with all GPU's on WCG, and there's going to be one he** of a demand. The # of CPUs here far exceeds the # of GPUs and there is plenty of work for them. I don't foresee the # of GPUs to be an issue with supply even with the faster speeds. Never happened on the HCC GPU project. There is an important difference from back then. This can run on integrated GPUs, which results in a huge increase in the number of GPUs that can run these work units. Even though they are much slower than discrete cards, they are much faster than CPUs. My RTX 2060 gets ~500 jobs/hr with only 26% load - the iGPU gets ~66 jobs/hr. That iGPU gets ~11x the jobs/hr vs all 8 threads in the i7 that it is a part of(tasks only on iGPU vs only on CPU). Whether that will result in a supply issue remains to be seen. If it does, that is a good problem to have.  |
||
|
|
|