Message boards :
Number crunching :
I've Built a Couple OSX CUDA Apps...
Message board moderation
Previous · 1 . . . 21 · 22 · 23 · 24 · 25 · 26 · 27 . . . 58 · Next
Author | Message |
---|---|
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
First, no such thing as a "100% Accurate App", because of the way floating point arithmetic, cross platform differences, and Cosmic rays work. (The last obscure item being one main reason real Mac Pros and other workstations have ECC memory, so do top end compute cards.) Looking at the result, you glitched out on spikes (for whatever reason). This could range from an actual system problem or instability, voltages/temps etc, to alpha particle emitting carbon in the packaging of your semiconductors (they all have some, nothing to do but hope it doesn't happen often) The trick is to see if you get more such 'glitching' then you have ammunition to check for simple possible problems to isolate and eliminate where possible. One off or intermittent events can be the trickiest to track down (where necessary). "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
I see a few more Tasks on his machine that are Very Suspect, http://setiathome.berkeley.edu/results.php?hostid=7952666&state=3 I just switched my Mac over to 10.11.4 with the CUDA 42 App to see if it shows the same behavior. As soon as the webpages catch up to reality we'll be able to see the results. The 10.11.4 Update affected either the OpenCL App, His machine, the CUDA 42 App, or any combination of the three. I'm pretty sure the Update affected his machine running the OpenCL App, it also appears the CUDA App isn't working right as well. The CUDA 42 App worked fine on my machine with 10.11.3, so, we'll see how it does with 10.11.4. It shouldn't take long, then I'm going to try the -poll command for about an hour. |
TimeLord04 Send message Joined: 9 Mar 06 Posts: 21140 Credit: 33,933,039 RAC: 23 |
TBar, UNPKG WILL NOT run on my system. After DL it and unzipping it, the App has a Prohibited sign over the App. Any other tool that will work in 10.11.4??? TL TimeLord04 Have TARDIS, will travel... Come along K-9! Join Calm Chaos |
TimeLord04 Send message Joined: 9 Mar 06 Posts: 21140 Credit: 33,933,039 RAC: 23 |
My personal intent was never to make you feel attacked, and the whole "we've been doing this for nearly 20 years thing" was to merely reinforce and hopefully get you to acknowledge that your machine was not acting appropriately with the th OpenCL version of the App, in spite of your repeated instance that it was. It was a fairly common problem that had a solution and for which we all offered you help to resolved. Thank you for the apology; it is most humbly accepted. :-) TL TimeLord04 Have TARDIS, will travel... Come along K-9! Join Calm Chaos |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
TBar, Did you try the App? If you get the warning about the Developer just hold down the control key while you Click on the App and tell it to Open. Do this first. Right click on unpkg.app, select Show Package Contents, then go to unpkg.app/Contents/Resources/appIcon.icns and open it. Tell me what you see on the Icon. |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
I see a few more Tasks on his machine that are Very Suspect, http://setiathome.berkeley.edu/results.php?hostid=7952666&state=3 Hopefully will get a chance to rationalise the 3 environments here this week as well. The Mac Pro here is still on 10.11.2. I don't really see how or why there would be changes there affecting Cuda, *Unless*, there is some sortof preparation for Pascal, Vulkan over Metal (which is a thing) or VR going on. "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
TimeLord04 Send message Joined: 9 Mar 06 Posts: 21140 Credit: 33,933,039 RAC: 23 |
TBar, OK - It worked. I have the two CUDA 65 Library Files. I copied and pasted them to the Desktop for now. Then I copied and pasted the other CUDA 65 Files from your App to the Desktop. I will set SETI@Home on Andromeda to No New Tasks, and run the queue down. Once the queue is empty, I will make the change to CUDA 65. Thank you for the directions. :-) [EDIT:] Just to confirm, I take the 6 Files that are now on my Desktop, and move them to the SETI@Home Project Folder. Do I reinstall BOINC at that point like I had to for CUDA 42??? TL [EDIT 2:] It's after 9 PM - Pacific, now... Have to wait until the queue is emptied sometime tomorrow. Then I'll make the changes to CUDA 65. TimeLord04 Have TARDIS, will travel... Come along K-9! Join Calm Chaos |
Wiggo Send message Joined: 24 Jan 00 Posts: 36584 Credit: 261,360,520 RAC: 489 |
Hopefully will get a chance to rationalise the 3 environments here this week as well. The Mac Pro here is still on 10.11.2. I don't really see how or why there would be changes there affecting Cuda, *Unless*, there is some sortof preparation for Pascal, Vulkan over Metal (which is a thing) or VR going on. Well I hope that will fix all those awful Mac's out there that seem to only want to throw invalids all over the place. I use to hate being winged by AMD GPU's and those with bad GTX 295's or 560/Ti's (the AMD GPU prob seems to be finally gone by the looks of it), but now Mac's are at the top of my list for the worst in reliability by producing crap results by the tanker load. :-( Cheers. |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Hopefully will get a chance to rationalise the 3 environments here this week as well. The Mac Pro here is still on 10.11.2. I don't really see how or why there would be changes there affecting Cuda, *Unless*, there is some sortof preparation for Pascal, Vulkan over Metal (which is a thing) or VR going on. Yeah no Idea on the OpenCL App on Mac. Was Running like crap on whatever I have now, so stopped it (since I need to use the machine sometimes) "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
OK - It worked. I have the two CUDA 65 Library Files. I copied and pasted them to the Desktop for now. Then I copied and pasted the other CUDA 65 Files from your App to the Desktop. I will set SETI@Home on Andromeda to No New Tasks, and run the queue down. Once the queue is empty, I will make the change to CUDA 65. Well, I've been running the CUDA 42 App with 10.11.4 for about 6 hours now. Not only do I Not see any suspicious Inconclusives, I Don't see any CUDA 42 Inconclusives at all. None, http://setiathome.berkeley.edu/results.php?hostid=6796479&state=3 In fact, almost of half of the 6th is gone and I only see 1 Inconclusive from the 6th, and that was from the "Special" CUDA 65 App. I see pages of tasks completed on the 6th but No Inconclusives. It would appear the App works as well today as it did over 2.5 months ago when it was posted. It would seem the 10.11.4 Update not only broke your Hackintosh with the OpenCL App but also the CUDA App, though not as badly with the CUDA App. I don't expect to see any change with the CUDA 65 App. Your best bet is to go back to 10.11.3 which seemed to work with the OpenCL App. That will probably solve the issue with the CUDA App as well. The CUDA 65 App will be a little faster but I wouldn't expect it to be any different otherwise. The files you need are; app_info.xml MBv8_8.05r3344_sse41_x86_64-apple-darwin setiathome_x41zi_x86_64-apple-darwin_cuda65 libcudart.6.5.dylib libcufft.6.5.dylib If you change the Plan Class on the New app_info to match the Old app_info you could make the change anytime. It wouldn't cause any problems to run the CUDA 65 App with the CUDA42 Plan class. What matters is that the Existing Tasks in the State file have the same Version Number and Plan Class as the app_info. But, it wouldn't hurt to run the tasks down either considering you already have over 60 Ghosts. Any time you edit/add a file in the setiathome.berkeley.edu folder you have to set the Permissions. The easiest way to set the correct Permissions is to just run the BOINC installer. The other ways involve the Terminal and Command Lines. Dang, where'd my inconclusives go? Is the Server working? |
TimeLord04 Send message Joined: 9 Mar 06 Posts: 21140 Credit: 33,933,039 RAC: 23 |
OK - It worked. I have the two CUDA 65 Library Files. I copied and pasted them to the Desktop for now. Then I copied and pasted the other CUDA 65 Files from your App to the Desktop. I will set SETI@Home on Andromeda to No New Tasks, and run the queue down. Once the queue is empty, I will make the change to CUDA 65. So, the "Libraries and Driver.rtf" file from the CUDA 65 App Folder is NOT needed??? I'll leave that out, then. I just copied all the files from the CUDA 65 App Folder to the Desktop, and that was one of them. TL TimeLord04 Have TARDIS, will travel... Come along K-9! Join Calm Chaos |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14676 Credit: 200,643,578 RAC: 874 |
"Libraries and Driver.rtf" is likely to be a documentation file in Rich Text Format - probably of more use to developers than end-users. Interesting, but not needed for crunching, no. |
TimeLord04 Send message Joined: 9 Mar 06 Posts: 21140 Credit: 33,933,039 RAC: 23 |
"Libraries and Driver.rtf" is likely to be a documentation file in Rich Text Format - probably of more use to developers than end-users. Interesting, but not needed for crunching, no. Thanks Richard; I'll just leave it out, then. :-) TL TimeLord04 Have TARDIS, will travel... Come along K-9! Join Calm Chaos |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Oh well. Run the "Baseline" CUDA App for a day and the Inconclusives are almost gone, I ended up with 3 against obviously bad wingpeople. Go back to the "Special" CUDA App and the Inconclusives are back. As mentioned before, there are reasons you still can't download the "Special" App, although it is better than it was a couple months ago. I find nothing wrong with the "Baseline" Apps running 10.11.4, they still work Great and should be on Beta being tested by all those troublesome Apple LapTops. When I run the "Baseline" Apps with the -poll command not only are the Inconclusive results almost gone, but the times are the fastest I can find on Beta for a 750Ti. Typical Shorty; http://setiathome.berkeley.edu/result.php?resultid=4840388593 Typical 0.42; http://setiathome.berkeley.edu/result.php?resultid=4840187676 Those are times for a Reference clocked 750Ti, 1020 MHz. |
TimeLord04 Send message Joined: 9 Mar 06 Posts: 21140 Credit: 33,933,039 RAC: 23 |
Update: Just finished all the CUDA 42 Tasks, and Upgraded to CUDA 65. Reinstalled BOINC 7.6.22 for good measure. Am now running CUDA 65 Tasks in 10.11.4. Let's see how this works. ;-) :-) TL [EDIT:] Also, still crunching 2 Units at a time. TimeLord04 Have TARDIS, will travel... Come along K-9! Join Calm Chaos |
TimeLord04 Send message Joined: 9 Mar 06 Posts: 21140 Credit: 33,933,039 RAC: 23 |
Well, CUDA 65 has been crunching 2 WUs a time for a few hours now; since around 8-ish Pacific, and Inconclusives have dropped to 22, Invalids remain at 3, (from CUDA 42), and of the 22 Inconclusives 7 are from CUDA 42. I think CUDA 65 may be the answer for my system. I will keep watching; and I'm sure that TBar will be watching as well. :-) TL TimeLord04 Have TARDIS, will travel... Come along K-9! Join Calm Chaos |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Presuming it's my Cuda65 build, some amount of variance in TBar's Cuda42 could be an artifact of TBar using the old Makefile system. Probably doesn't account for all of whatever was going on, but some prior compiler options used for speed are no longer precise enough for tight Seti v8 match. Not a huge concern, though does push the build system rationalisation issue up the priority list (3 different buildsystems is not sustainable) "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
TimeLord04 Send message Joined: 9 Mar 06 Posts: 21140 Credit: 33,933,039 RAC: 23 |
Just got 2 Inconclusives on CUDA 65. CUDA 65, First 2 Units in List. Both Reported 4-7-2016. So far, still lower in Inconclusives than CUDA 42, and MUCH lower in Inconclusives than OpenCL. TL TimeLord04 Have TARDIS, will travel... Come along K-9! Join Calm Chaos |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Just got 2 Inconclusives on CUDA 65. You're still seemingly glitching out on spikes there. Your diagnostic options are pretty limited on Mac. Presuming there is still no clock or fan control, you may be limited to hoping a driver reinstall cleans up something. is nvidia-smi available via the command line ? [Edit:] note that in one of the two cases, probably the wingman isn't that great, will have to give it some time and see. The one spike glitch task we'll see if happens again. A once off event could be anything. "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
TimeLord04 Send message Joined: 9 Mar 06 Posts: 21140 Credit: 33,933,039 RAC: 23 |
Just got 2 Inconclusives on CUDA 65. No clock or fan control, even 3rd Party that I can find. I can reinstall the Web Driver, 346.03.06f01, and CUDA Driver, 7.5.26 again... If it will help. I will wait for now, though. [EDIT:] I don't know what nvidia-smi is, or does. I don't make use of Command Lines. TL TimeLord04 Have TARDIS, will travel... Come along K-9! Join Calm Chaos |
©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.