| 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 26 2021 [ Issues Thread ] |
| No member browsing this thread |
|
Thread Status: Locked Total posts in this thread: 511
|
|
| Author |
|
|
bozz4science
Advanced Cruncher Germany Joined: May 3, 2020 Post Count: 104 Status: Offline Project Badges: 
|
I don't remeber having seen my GPU VRAM spike even that high with 3 WUs computed concurrently on each. Likely other factors that might restrict performance of those cards, other than VRAM (IMO). You should however wait for other people's feedback on your question that might be able to better help you out here!
---------------------------------------- AMD Ryzen 3700X @ 4.0 GHz / GTX1660S Intel i5-4278U CPU @ 2.60GHz |
||
|
|
bozz4science
Advanced Cruncher Germany Joined: May 3, 2020 Post Count: 104 Status: Offline Project Badges:
|
@uplinger: Any word from the researchers at Scripp how they currently plan the schedule of the GPU app rollout out of the current beta?
----------------------------------------AMD Ryzen 3700X @ 4.0 GHz / GTX1660S Intel i5-4278U CPU @ 2.60GHz |
||
|
|
Richard Haselgrove
Senior Cruncher United Kingdom Joined: Feb 19, 2021 Post Count: 360 Status: Offline Project Badges: 
|
I wonder why, after 100 pages of comments, I don't remember reading anything on hardware requirements. I've been running some instrumented offline tests because of repeated failures with my Intel HD 4600 iGPU. I can say that (from a GPU-Z sensor log), the highest dedicated memory usage I've recorded is 213 MB - so memory wouldn't seem to be a significant constraint.The problems I've been encountering seem to be concerned with timing: the slowest cards may have problems. But I'll be writing privately to Keith about that later today. |
||
|
|
Grumpy Swede
Master Cruncher Svíþjóð Joined: Apr 10, 2020 Post Count: 2068 Status: Recently Active Project Badges: 
|
I think the app does checkpoint between each Autodock component. On a fast discrete GPU (NVidia or AMD), that's barely worth it. But on an Intel iGPU, a full Beta run takes up to 40 minutes, and may last longer if more complex molecules are studied after the final application is released from Beta. If people are worried about additional disk writes on a SSD (for example), they can extend BOINC's interval between checkpoints to longer than the typical runtime on their specific GPU. Having said that, I think the largest disk writes are: 1) The initial unpacking of the task data before the run fully starts 2) Adding result data to the output files for return to WCG Suppressing checkpoints isn't going to make much difference to either of those essential procedures. I have my checkpoints (write to disk at most every X seconds) set to 1800 seconds, but several WCG projects does not honour that setting at all. This GPU beta for example, checkpoints in between every "job", and as I said, it does not honour the "write do disk....." setting. That's a lot of writing to disk if the WU contains 30 or more "jobs", and the whole WU takes only 2 minutes to crunch. Edit, added: I think I said early on in this Beta, that the app had been writing to disk 5-6% of the time or something like that, after I had done some 40 tasks or so. [Edit 1 times, last edit by Grumpy Swede at Apr 4, 2021 3:02:10 PM] |
||
|
|
Crystal Pellet
Veteran Cruncher Joined: May 21, 2008 Post Count: 1313 Status: Offline Project Badges: 
|
If people are worried about additional disk writes on a SSD (for example), they can extend BOINC's interval between checkpoints to longer than the typical runtime on their specific GPU. I have my checkpoints (write to disk at most every X seconds) set to 1800 seconds, but several WCG projects does not honour that setting at all. This GPU beta for example, checkpoints in between every "job", and as I said, it does not honour the "write do disk....." setting. Grumpy is right. BOINC's setting 'write to disk every .. seconds' is ignored. My setting 'WtD' was every 120 seconds. I got a BETA and enable checkpoint_debug after the 8th job. 04 Apr 20:28:57 [checkpoint] result BETA_OPNG_0000267_00203_2 checkpointed 3 CPU OPN 's and this OPNG. 4 threads. i5-4310U CPU with HD Graphics 4400  |
||
|
|
ThreadRipper
Veteran Cruncher Sweden Joined: Apr 26, 2007 Post Count: 1317 Status: Offline Project Badges: 
|
If you have enough RAM, then a RAMDisk should save the SSD from checkpoints. I run ImDisk which is free and it has worked flawlessly as long as you don't choose A: or B: as you device letter for the RAMDisk Drive in Windows.
---------------------------------------- Join The International Team: https://www.worldcommunitygrid.org/team/viewTeamInfo.do?teamId=CK9RP1BKX1 AMD TR2990WX @ PBO, 64GB Quad 3200MHz 14-17-17-17-1T, RX6900XT @ Stock AMD 3800X @ PBO AMD 2700X @ 4GHz [Edit 1 times, last edit by ThreadRipper at Apr 4, 2021 7:06:17 PM] |
||
|
|
Quadophile
Cruncher USA Joined: Feb 4, 2006 Post Count: 25 Status: Offline Project Badges: 
|
If you have enough RAM, then a RAMDisk should save the SSD from checkpoints. I run ImDisk which is free and it has worked flawlessly as long as you don't choose A: or B: as you device letter for the RAMDisk Drive in Windows. You mean not assigning a drive letter to ramdisk?  |
||
|
|
robertmiles
Senior Cruncher US Joined: Apr 16, 2008 Post Count: 443 Status: Offline Project Badges:
|
If you have enough RAM, then a RAMDisk should save the SSD from checkpoints. I run ImDisk which is free and it has worked flawlessly as long as you don't choose A: or B: as you device letter for the RAMDisk Drive in Windows. You mean not assigning a drive letter to ramdisk? He probably means do assigning A: or B: as the drive letter, but still assigning some other drive letter. Those two drive letters used to be reserved for floppy disks, and some software still treats them that way. |
||
|
|
Quadophile
Cruncher USA Joined: Feb 4, 2006 Post Count: 25 Status: Offline Project Badges:
|
If you have enough RAM, then a RAMDisk should save the SSD from checkpoints. I run ImDisk which is free and it has worked flawlessly as long as you don't choose A: or B: as you device letter for the RAMDisk Drive in Windows. You mean not assigning a drive letter to ramdisk? He probably means do assigning A: or B: as the drive letter, but still assigning some other drive letter. Those two drive letters used to be reserved for floppy disks, and some software still treats them that way. That makes sense. Thank you for your input  |
||
|
|
hnapel
Advanced Cruncher Netherlands Joined: Nov 17, 2004 Post Count: 82 Status: Offline Project Badges:
|
How do you do this? Some special BOINC setting? My BOINC client is only running as much GPU jobs as there are GPUs (2x Nvidia, 1x Intel HD) at the same time. Also as it was explained the GPU jobs will get more complicated over time with the (apparent) subunits taking more time so the problem would ease out. Why would you want more GPU jobs to run at the same time on one GPU, this would be different from the CPU jobs where you always run one WU per core (or thread in case of hyperthreading). You run multiple GPU tasks at the same time by creating an app_config file with the correct syntax and placing it into your WCG project folder. You then go to the options tab in BOINC and select "read config files". A search here should give you all the info you need.Thanks, but I only want it of the NVIDIA GPUs, so this should use the plan_class directive to separate the NVIDIA from Intel HD graphics, so what is the plan_class for these jobs for the Nvidia? An example would be nice. |
||
|
|
|