Message boards :
Number crunching :
SETI@home v8.22 Windows GPU applications support thread
Message board moderation
Previous · 1 . . . 10 · 11 · 12 · 13 · 14 · 15 · 16 . . . 23 · Next
Author | Message |
---|---|
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
Ok, run your GPU cache down, then move the old r3557 aistub to the oldApp_backup folder, edit the newer aistub to reflect the correct entries and then run the aimerge.cmd again, that should fix it. . . After restoring the correct version number to the r3557 entry I rebooted and it looked OK until I tried to get work, then most of the job types (CUDA50, SoG and something else) all said there was no app associated with them, even though they were pointed to r3584 and the files are all there. It all beats me ... so I re-ran Lunatics to get things working again. Stephen [edit] That was a while ago :) ? ? |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13161 Credit: 1,160,866,277 RAC: 1,873 |
I said in an earlier post that aimerge will include ALL .aistub files into the app_info it generates and that you have to remove the older .aistub files you don't want to include. That means any .aistub files for AP work too if those are in the directory. If you don't want to include ALL the plan_classes that Richard put into the Lunatics installer, just edit the .aistub file to remove the extraneous older fluff. That is what I did and also what I suggested you do also earlier. So you should have used only the .aistub file from the r3584 archive to run aimerge against. That is the corrected .aistub file I posted with the missing .CL file added. Any time you want to change a plan_class or version level, you MUST run down your cache to nil before making the change. It has been that way since the beginning. Nothing new there. Again, I will point out that what the task is labelled as in your cache is meaningless. I still have my gpu tasks labelled as 8.20 even though I am using the r3584 application. I could care less what they are labelled as. They still validate and award the same credit whether they are labelled 8.20 or 8.22. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
I said in an earlier post that aimerge will include ALL .aistub files into the app_info it generates and that you have to remove the older .aistub files you don't want to include. That means any .aistub files for AP work too if those are in the directory. If you don't want to include ALL the plan_classes that Richard put into the Lunatics installer, just edit the .aistub file to remove the extraneous older fluff. That is what I did and also what I suggested you do also earlier. . . The problem is I am crunching on the CPU and on this rig I do crunch AP work so I need Richard's aistub. But I guess the only way to get it to work is to do some heavy pruning on it and hope it will get it right. Stephen :( |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13161 Credit: 1,160,866,277 RAC: 1,873 |
. . The problem is I am crunching on the CPU and on this rig I do crunch AP work so I need Richard's aistub. But I guess the only way to get it to work is to do some heavy pruning on it and hope it will get it right. It isn't that hard. Just snip delimiter sections out that you don't want. You just have to make sure any entry is delimited on each side by the [x] [/x] entry. For a sanity check, open the file with a browser as .XML and let it pick up any syntax errors. Gross errors are always caught but multiple embedded errors can slip by occasionally. It isn't that hard to read the file looking for proper callouts and use of proper syntax. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
. . The problem is I am crunching on the CPU and on this rig I do crunch AP work so I need Richard's aistub. But I guess the only way to get it to work is to do some heavy pruning on it and hope it will get it right. . . I would feel more comfortable if it has nice big section markers like a line of tilde's or such across the screen between sections :) . . I have copied all the aistubs from my projects folder into a work folder to look at them all and sort out any conflicts, or overlaps. Hopefully I can come up with something small, tidy and right. The problem is, even with the sanity check, I am not sure I can be confident of getting it right ... Stephen <shrug> |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13161 Credit: 1,160,866,277 RAC: 1,873 |
If you still have doubts, then just work your cache down to a dozen tasks or so including at least one of each type. Then if you make a mistake in editing you only dump the dozen. I have never dumped any work once I learned the syntax structure of the .aistubs and the app_info. I just make sure the syntax is correct in the first place. There are only half a dozen structures in any file. They were not that hard to figure out. And you always have the BOINC wiki's to fall back on for examples. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Miklos M. Send message Joined: 5 May 99 Posts: 955 Credit: 136,115,648 RAC: 73 |
Just as I am getting happy with all those 8.22 wu's, I started getting again "50's". Nothing was done to my computer. I just went to sleep for the night and it was running all night. I wonder if anyone could tell me what is wrong here? Thank you |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
Just as I am getting happy with all those 8.22 wu's, I started getting again "50's". Nothing was done to my computer. I just went to sleep for the night and it was running all night. I wonder if anyone could tell me what is wrong here? . . It looks like the servers think you are back to stock. You have 255 CUDA50 and CUDA42 tasks that you have aborted. . . Check that your app_info.xml file is still intact. Stephen ? ? |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
If you still have doubts, then just work your cache down to a dozen tasks or so including at least one of each type. Then if you make a mistake in editing you only dump the dozen. . . Hospital tomorrow, so this will wait till the weekend. Stephen :( |
Miklos M. Send message Joined: 5 May 99 Posts: 955 Credit: 136,115,648 RAC: 73 |
Stephen, please tell me step by step( I do not know computers, I just like to crunch numbers) how do I get to app_info.xml file. Thank you |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
Stephen, please tell me step by step( I do not know computers, I just like to crunch numbers) how do I get to app_info.xml file. . . Open File Explorer and find your BOINC directory (use the search function) then select that directory. If you open the tree (the little arrow head will point down instead of to the right) you will see subfolders, select ProgramData, then BOINC, then projects. This will have folders for each project you are attached to. Select setiathome.berkeley.edu. There are a lot of files in this folder but if you scroll through them to app_info_xml then you are in business. . . The hard part is knowing what to make of the contents of this very large file, but you will need to use Notepad to open it by right clicking on it and choosing "Open with" then notepad. . . If the file is not in that folder then you may need to re-run Lunatics Installer, but that will take you back to version r3557 (v8.20). . . I hope that helps. Stephen . |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13161 Credit: 1,160,866,277 RAC: 1,873 |
He is running stock apps, not Lunatics. You are getting CUDA50 and CUDA42 tasks AS DESIGNED by the servers along with the SoG tasks. Nothing you can do about it. This is normal for the servers to send every type of task to find which is most efficient on your systems. The only way to run exclusive SoG tasks is to write your own app_info or use the Lunatics installer to make a Anonymous platform. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Cactus Bob Send message Joined: 19 May 99 Posts: 209 Credit: 10,924,287 RAC: 29 |
I see a lot of people use notepad to edited files and for the most part it works but has a few easy pitfalls, its very easy to save to the wrong file and it has no syntax checking just for starters. I use Atom Editor to change any files needed for BOINC. It has format, syntax ad spell checking to help you see errors easier. It's free and there a TON of videos and web pages out there to help you make Atom work for you. https://atom.io/ Atom Site https://en.wikipedia.org/wiki/Atom_(text_editor) Atom Wiki I think there a couple other free editors out there that may be as good but I haven't used them. I just know Atom works much better than a word type text editor. If this has been mentioned before than sorry for the spam. Bob Sometimes I wonder, what happened to all the people I gave directions to? ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ SETI@home classic workunits 4,321 SETI@home classic CPU time 22,169 hours |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13161 Credit: 1,160,866,277 RAC: 1,873 |
Never heard of it before. Looks interesting. I have been using Notepad++ for a while for my text file editing. Lots of features too. I'm downloading Atom as I type to look at it and see if it can be useful in syntax checking. [Edit] Lots of features. Will take a long time to learn how to use it. The goto matching bracket feature is the most obvious feature at first glance useful to someone editing a BOINC XML file. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Miklos M. Send message Joined: 5 May 99 Posts: 955 Credit: 136,115,648 RAC: 73 |
Thank you Stephen. Followed your directions since I could not find an infoxml in my folders. Downloaded Mike's and now I got 8.20 wu's with ocl. very happy again. Miklos |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
. . Thanks, I'll have a look ... Stephen . |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
Thank you Stephen. Followed your directions since I could not find an infoxml in my folders. Downloaded Mike's and now I got 8.20 wu's with ocl. very happy again. . . I am happy to help. Stephen :) |
iwazaru Send message Joined: 31 Oct 99 Posts: 173 Credit: 509,430 RAC: 0 |
Yeah, nobody has really cracked the correlation between Nvidia card specs and SETI throughput. CUs & memory bandwidth are the best indicator but mostly for same gen silicon (and even those metrics don't scale too well). Things get messier when you try to compare Pascal gen by going all the way back to Fermi gen... One thing to remember is that Nvidia played a little switcheroo on us. Back in the Golden Days of Seti Credit a 560Ti (332mm2, xx4 class) cost $249. Today, that kind of silicon will cost you $599 since a 314mm2, xx4 class card is now called a GTX 1080 :) But yeah, Compute Units is what we should be looking at and not Cuda "cores" (because they aren't really cores). And those compute units are Memory Bandwidth constrained so unless there's another bottleneck somewhere we don't know yet (Seti-wise that is) then HBM should give a slight boost to our apps whenever it becomes mainstream. Shaggie's seminal GPU FLOPs charts are by far the best way to make heads or tails of the pecking order though. |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13161 Credit: 1,160,866,277 RAC: 1,873 |
Yes it is the CU or SM (Shader Module) that is the complete compute engine that does the calculations. So the more shaders, the more processing a card can do. And I would want to see actual data showing that HBM memory is faster than GDDR5/5X memory when it comes to compute bandwidth. The miners have pretty much disavowed any GDDR5X or HBM memory configured cards. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13722 Credit: 208,696,464 RAC: 304 |
And I would want to see actual data showing that HBM memory is faster than GDDR5/5X memory when it comes to compute bandwidth. The miners have pretty much disavowed any GDDR5X or HBM memory configured cards. Like CPU cache & CPU to system memory bandwidth- if your application requires it, then it makes a big difference. If it doesn't require much in the way of memory bandwidth, then not having it doesn't matter. I'm pretty sure Raistmer mentioned in one of his posts on SoG development that GPU memory bandwidth was a significant factor impacting on GPU performance. Grant Darwin NT |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.