| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Active Research
Forum: OpenPandemics - COVID-19 Project
Thread: OpenPandemics - GPU Stress Test |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 781
|
|
| Author |
|
|
nanoprobe
Master Cruncher Classified Joined: Aug 29, 2008 Post Count: 2998 Status: Offline Project Badges: 
|
and actually, bigger tasks is what they're pumping out with these high batch number tasks. anything 13,000+ it seems. they're running up to 10x longer than the 7000 and under batches on my 2080ti. (13,000+ = ~9 minutes, 7000- = ~1min or less) none have validated yet (pending), the question is, will they receive 10x credit as the older tasks? my gut feeling says no. uplinger has mentioned in a previous posts that the WUs have some maximum credit reward baked in, and that actual credit granted is based on the percentage of calculations needed to get the result. NOT the total calculations. and both old and new tasks are tagged with the same estimated flops size (incorrectly) at 31,450 GFlops, so my guess is that the maximum credit is the same as well. Max points per task is 1800. The two validated 5 digit tasks I had on 4/25 were given high 1600s low 1700s.
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.
  |
||
|
|
squid
Advanced Cruncher Germany Joined: May 15, 2020 Post Count: 56 Status: Offline Project Badges: 
|
A load balancer intelligently distributes traffic from clients across multiple servers without the clients having to understand how many servers are in use or how they are configured.
https://kemptechnologies.com/de/load-balancing/http-load-balancer/ |
||
|
|
Ian-n-Steve C.
Senior Cruncher United States Joined: May 15, 2020 Post Count: 180 Status: Offline Project Badges: 
|
and actually, bigger tasks is what they're pumping out with these high batch number tasks. anything 13,000+ it seems. they're running up to 10x longer than the 7000 and under batches on my 2080ti. (13,000+ = ~9 minutes, 7000- = ~1min or less) none have validated yet (pending), the question is, will they receive 10x credit as the older tasks? my gut feeling says no. uplinger has mentioned in a previous posts that the WUs have some maximum credit reward baked in, and that actual credit granted is based on the percentage of calculations needed to get the result. NOT the total calculations. and both old and new tasks are tagged with the same estimated flops size (incorrectly) at 31,450 GFlops, so my guess is that the maximum credit is the same as well. Max points per task is 1800. The two validated 5 digit tasks I had on 4/25 were given high 1600s low 1700s. That validates my prediction. Thank you.  EPYC 7V12 / [5] RTX A4000 EPYC 7B12 / [5] RTX 3080Ti + [2] RTX 2080Ti EPYC 7B12 / [6] RTX 3070Ti + [2] RTX 3060 [2] EPYC 7642 / [2] RTX 2080Ti |
||
|
|
nanoprobe
Master Cruncher Classified Joined: Aug 29, 2008 Post Count: 2998 Status: Offline Project Badges:
|
Yup batches 13345 - 41773 seems to be of the larger type. More "jobs"/WU, and maybe more complicated "jobs". They do take considerably longer time to crunch though, and seems to take some time at the start, before they really begin crunching. Meaning the percentage starts rising from the beginning, but that's BOINC's pseudo-progress, then after some time, it backs down to the real percentage, after the first "job" is done. I haven't seen that on the previous batches, other than on my really slow GPU's (GTX660M, and iGPU HD4600) However, no peace of mind, for those with BOINC on SSD's. They still hammer the disk between each "job", when they checkpoint. I have BOINC on a HD, so I'm not that worried. All these tasks start out on the CPU first, usually 20 seconds or less. These 5 digit batches use far more CPU time at startup and that is the progress you see before the task is handed off to the GPU. They also use far more CPU time for the entire running of the tasks.
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.
|
||
|
|
biini
Senior Cruncher Finland Joined: Jan 25, 2007 Post Count: 334 Status: Offline Project Badges: 
|
"c:\program files\boinc\boinccmd.exe" --network_available Then add loop/repeat controls as appropriate to your desires and scripting language. Thanks for the tip! does that also retry stalled transfers ? |
||
|
|
maeax
Advanced Cruncher Joined: May 2, 2007 Post Count: 142 Status: Offline Project Badges: 
|
One Ellesmere with HDD and one Ellesmere with SSD. The SSD one is crashing very often, because of a lot Checkpoints. Boinc is set to 1200 sec. for backup, but OPNG ignore this. Now are the longrunning OPNG-Tasks running on it (1 hour!). Something is wrong with checkpointing and SSD. https://www.worldcommunitygrid.org/ms/device/...s.do?workunitId=639284992 Either something is wrong with your SSD, or something else is wrong with the system with the SSD. My systems are much faster and running 6-8 GPUs and producing many more writes to the SSD, but with no issues. SSDs in general are capable of many orders of magnitude more IOPs than a HDD. Your problem is likely system-specific, not SSD-specific. No problem with Einstein@Home!
AMD Ryzen Threadripper PRO 3995WX 64-Cores/ AMD Radeon (TM) Pro W6600. OS Win11pro
|
||
|
|
spRocket
Senior Cruncher Joined: Mar 25, 2020 Post Count: 274 Status: Offline Project Badges:
|
Things seem to have settled down here, and I'm seeing 15xxx WUs now, they're taking about 15 minutes for me.
|
||
|
|
m0320174
Cruncher Joined: Feb 13, 2021 Post Count: 11 Status: Offline Project Badges: 
|
I'm able to grasp enough CPU work but I'm continuously running out of GPU workunits. When I fetch for work I usually get 0, 1 or 2 GPU workunits, not enough to keep my GPU busy (I complete them in +/- 2 minutes). I don't know - I've been getting 16-20 WUs when BOINC decides to report completed work. Could this depend on Intel/Nvidia/AMD? In any case, the stuck uploads don't seem to be affecting my ability to fetch work, just to return it. The returns eventually make it, but it's kind of amusing to see one half of an OPNG unit upload fine and the other half get stuck. I'm wondering if OPNG units need to be made a bit bigger? Mine generally take 3-5 minutes on my GTX 960. bigger tasks will certainly help the situation. bigger tasks = longer runtime = less frequent work requests = less server load. And the bigger tasks indeed resolve my issue. I'm no longer running out of GPU work. It also seems to have a positive effect on the upload/download server (might be a coincidence). |
||
|
|
uplinger
Former World Community Grid Tech Joined: May 23, 2005 Post Count: 3952 Status: Offline Project Badges: 
|
Good morning,
I should have posted sooner, since I've been up for 2 hours. We are still working on tweaking some of the load balancer values. It seems like things are running a bit smoother right now. We are only in about 20 minutes on the latest changes. Please let us know if you are noticing anything on your end. Thanks, -Uplinger |
||
|
|
Jim1348
Veteran Cruncher USA Joined: Jul 13, 2009 Post Count: 1066 Status: Offline Project Badges: 
|
Please let us know if you are noticing anything on your end. It is much better now. All my stuck OPNG uploads have gone, as well as the MCM and MIP ones. There are still a number of "download errors" on new MIPs, but I am sure that will clear up shortly. I will lay off OPNG for a while to let you get your breath. |
||
|
|
|