Message boards :
Number crunching :
Linux CUDA 'Special' App finally available, featuring Low CPU use
Message board moderation
Previous · 1 . . . 59 · 60 · 61 · 62 · 63 · 64 · 65 . . . 83 · Next
Author | Message |
---|---|
Jeff Buck Send message Joined: 11 Feb 00 Posts: 1441 Credit: 148,764,870 RAC: 0 |
Hi,Does that apply to zi3t2b as well? I have that version running on 2 of my Linux boxes, but have zi3v running on the other one. The reason I haven't moved all of them to zi3v is that annoying problem with restarted tasks spewing out phantom spikes or triplets after the restart until the task overflows, resulting in the task getting marked Invalid. About EDIT: Meant to say 15%, or about 3 per week out of 20 restarted tasks. |
petri33 Send message Joined: 6 Jun 02 Posts: 1668 Credit: 623,086,772 RAC: 156 |
Hi,Does that apply to zi3t2b as well? I have that version running on 2 of my Linux boxes, but have zi3v running on the other one. The reason I haven't moved all of them to zi3v is that annoying problem with restarted tasks spewing out phantom spikes or triplets after the restart until the task overflows, resulting in the task getting marked Invalid. About EDIT: Sorry. I'll check and TBar will chek. Check your code for t2b. If the pulse find code does not have C_SCAN then it should not be used. grep C_SCAN cuda/*.cu The command should find a lot of lines. If none is found it does not have sequential scan in it. I'll add the restart problem to my to do list. To overcome Heisenbergs: "You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones |
Jeff Buck Send message Joined: 11 Feb 00 Posts: 1441 Credit: 148,764,870 RAC: 0 |
As far as I know, the previously discovered Best Gaussian problem discovered with the Windows SoG App DOES cross validate, and STILL EXISTS. You don't seem very concerned about that problem, and it's actually more troublesome than an occasional race condition with the Best Pulse.It's not that problems like that don't still concern me, it's just that it's been pretty thoroughly discussed and enough examples of the issue have already been posted. I try to keep an eye out for issues that appear in the latest versions of the apps, whether they're new problems, or continuing ones that I would hope would have been fixed in those newer versions. The two Best Pulse examples that I posted are the first I've seen with the Cuda 9.0 app and, since I'm not the one running that version, they only show up to me when one of my wingmen is running it and then that WU shows up in my Inconclusives list. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
The single signal if I recall correctly where score depends not only from peak is Gaussian. In all other peak and score correlate monotonically.
Very bad point of view and approach in general. For result to be correct ALL signals should match in ~1% tolerance. No any "half to be wrong" at all! Where "half" appears is just credits awards to encourage users to continue participation, nothing more. Devs should never take this into account. We are here not to collect credits. If the score is less than a given threshold then it is reported as best so far just to make the screen saver happy and to make an educated guesses of a sequential apps inner workings. Nope. To check processing correctness even on relatively silent data sets. The is no scientific meaning in those not reported but best anyway still pulses. They are there to prevent faking. One could say that no pulses were found without scanning through all possibilities. The best but not reported is a sanity check. If my app fails that sometimes it is not so big a deal. So it has meaning. Yes, some sort of CRC for processing pipeline to be short. And I'm working on it. That's good part. Either processing logic for "best" in your app greatly differs from "reportable" or issue with best could show itself on reportable ones too. So better to explore this more.
That's usual issue with any open testing. Not all testers follow guidlines. And that's why we all agreed so far that Windows builds should be postponed. My SW does all the work needed. No faking. Everything is computed. Nobody suspect anything else AFAIK. But your code just as stock one is GPL so available. And nor you not anyone else could prevent to use it in malice way. So as baseline we should not neglect "CRC checking". The problem is in (storing intermediate results on same PoT) the reporting, my lack of time during the weeks I have to go to the work and the day having only 24 hours in it during the weekends. And that would require additional serialization perhaps. And any serialization quite slow on big GPUs. That will constitute an issue when you will face the need to replace "fast but wrong some time" to "rock-stable but a little slower". Especially taking into account that "half could be wrong" part and obsession with credits for some participants. SETI apps news We're not gonna fight them. We're gonna transcend them. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Hmmm, where has the problem with the SoG Best Gaussian been thoroughly discussed other than this thread? I must have missed that. The Best Pulse problem with the CUDA App has been run into the ground, go back to the first page of this thread and you will find it there. There is a difference though, we still don't know what causes the problem with the Rare Bad Best Pulse whereas we Do know the problem with the SoG OpenCL App and Bad Best Gaussian. Just use the Non-SoG OpenCL App and the Bad Best Gaussian goes away. We are still working on the Bad Best Pulse and fortunately it is rare and only effects around 100 people, it also rarely gets by the Validation phase. On the other hand, there are 1000s of times more people using the SoG App and it constantly slips by because other SoG Apps validate against it. The only time it shows itself is against other Apps such as the CPU only machines. Don't bother looking for it on an SoG machine, just look at a CPU machine, https://setiathome.berkeley.edu/workunit.php?wuid=2675920971As far as I know, the previously discovered Best Gaussian problem discovered with the Windows SoG App DOES cross validate, and STILL EXISTS. You don't seem very concerned about that problem, and it's actually more troublesome than an occasional race condition with the Best Pulse.It's not that problems like that don't still concern me, it's just that it's been pretty thoroughly discussed and enough examples of the issue have already been posted. I try to keep an eye out for issues that appear in the latest versions of the apps, whether they're new problems, or continuing ones that I would hope would have been fixed in those newer versions. The two Best Pulse examples that I posted are the first I've seen with the Cuda 9.0 app and, since I'm not the one running that version, they only show up to me when one of my wingmen is running it and then that WU shows up in my Inconclusives list. Best gaussian: peak=4.105192, mean=0.5110582, ChiSq=1.134106, time=39.43, d_freq=1420289942.89, score=2.552872, null_hyp=2.239423, chirp=-30.138, fft_len=16k Best gaussian: peak=3.76384, mean=0.5175309, ChiSq=1.381263, time=41.1, d_freq=1420289892.33, score=2.347487, null_hyp=2.354173, chirp=-30.138, fft_len=16k Do you have any idea how many of the SoG Tasks are Cross Validating with the Wrong Bad Best Gaussian? Let me put it this way, the Bad Best Pulse numbers Pale in comparison. But, we know the source of the Gaussian problem, it could be fixed very simply. However, it appears people have decided not to worry about all those Bad Best Gaussians and instead focus on the Rare CUDA Best Pulse. Some of us know the Score, I would suggest you consider the people that have decided all those Bad Best Gaussians are OK before giving Petri a hard time about a much less common problem. As he said, he is working on it. If others want to help they are more than welcome, it would be very easy to setup a nVidia machine with the Special CUDA App. Install the OS then add this package, http://www.arkayn.us/forum/index.php?topic=197.msg4515#msg4515 BTW Raistmer, do you think you could help convince Eric to switch the Mac ATI App over to the non-SoG version? As more people update to the new OS the number of Failed tasks are going to be impressive. |
mmonnin Send message Joined: 8 Jun 17 Posts: 58 Credit: 10,176,849 RAC: 0 |
Too much here to read through the whole thread when it goes on for several pages on how to install Ubuntu so please forgive it's been asked in the many pages in between. I already have a BOINC install I don't want to mess with. Can I copy over these files in from the below link to the SETI project directory that I have now? I have a single 1070 and when I compared with several of the top hosts they are producing a lot more per Run time/GPU. http://www.arkayn.us/forum/index.php?topic=197.msg4515#msg4515 |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
You just need to download the 3 parts of the package in the message and unpack them to your Desktop. Double-click on the BOINC client and you are up and running. Also the message states you need to download the CUDA 8.0 libraries from the location in the readme.txt file and put those files into the SETI project directory the package created. The package already has the necessary AP and MB files with the app_info for those applications already installed. Couldn't be more simple. You might just need to write your own app_config to tailor exactly how you want to run the tasks. [Edit] Read your post again and missed the existing BOINC install under Linux. You just need to add the special app, download the CUDA 8.0 libraries and rewrite your app_info to handle the special app declaration. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Jeff Buck Send message Joined: 11 Feb 00 Posts: 1441 Credit: 148,764,870 RAC: 0 |
Hmmm, where has the problem with the SoG Best Gaussian been thoroughly discussed other than this thread? I must have missed that. The Best Pulse problem with the CUDA App has been run into the ground,Why does it need to be discussed elsewhere? If any topic was run into the ground, I'd say it was the Best Gaussian. I know. I was pretty deeply involved in that back in June, spending a lot of time combing through my Inconclusives and posting many, many examples. But it seems to me that the ultimate conclusion was that it might not be exclusive to SoG. That, in fact, there was some divergence in the code paths that went back several years. As I recall, Jason had intended to dig into it some more but he's been kind of disengaged since then. Perhaps when he drifts back in, the discussion can pick up. In the meantime, unless something changes, it seems rather pointless to keep posting additional examples. And whether or not the Best Pulse problem with the CUDA app has been "run into the ground", the fact that a new version of the Special App has started showing up makes it eminently reasonable to take a close look at the results to see if anything has changed, been improved, or fixed. Simply because it's a few seconds faster doesn't mean we should all bow down, close our eyes and just give thanks. If a new SoG app had arrived, the same scrutiny should be applied. The difference there is that new SoG apps (and just about every other new app) hit Beta first, whereas the Special App just goes straight to Broadway. Hence, close scrutiny is even more important. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Really? Send me the patch then please. I'll add it to repo from old host. BTW Raistmer, do you think you could help convince Eric to switch the Mac ATI App over to the non-SoG version? As more people update to the new OS the number of Failed tasks are going to be impressive. Think it's possible. Just abandon some of plan classes for Mac. Post few such hosts please. SETI apps news We're not gonna fight them. We're gonna transcend them. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
When I tested it in Linux all I had to do was use the Non-SoG version to solve the Best Gaussian problem. If you look at the Apps page it shows Windows as the only platform that doesn't have a Non-SoG nVidia App. I would suggest you build a Non-SoG Windows App and see if it has the Gaussian problem, I'd wager it won't. You can see the Mac results at Beta, along with the posts in the Questions & Answers section, http://setiweb.ssl.berkeley.edu/beta/setiathome_v8_x86_64-apple-darwin__opencl_ati5_mac.html http://setiweb.ssl.berkeley.edu/beta/setiathome_v8_x86_64-apple-darwin__opencl_ati5_SoG_mac.html SoG OS 17.0.0 : hosts_success 0.5000 : results_success 0.6091 Non SoG OS 17.0.0 : hosts_success 1.0000 : results_success 0.8854 This Host just changed to the Non SoG App after posting in Q & A, https://setiathome.berkeley.edu/results.php?hostid=8248108&offset=40 |
-= Vyper =- Send message Joined: 5 Sep 99 Posts: 1652 Credit: 1,065,191,981 RAC: 2,537 |
Hmm What to do about this? Running Mint 17.3 with Nvidia 384.90.. "<core_client_version>7.2.42</core_client_version> <![CDATA[ <message> process exited with code 127 (0x7f, -129) </message> <stderr_txt> ../../projects/setiathome.berkeley.edu/setiathome_x41zi3xs2_x86_64-pc-linux-gnu_cuda90_static: relocation error: ../../projects/setiathome.berkeley.edu/setiathome_x41zi3xs2_x86_64-pc-linux-gnu_cuda90_static: symbol _ZTVNSt7__cxx1119basic_istringstreamIcSt11char_traitsIcESaIcEEE, version GLIBCXX_3.4.21 not defined in file libstdc++.so.6 with link time reference </stderr_txt> ]]> " _________________________________________________________________________ Addicted to SETI crunching! Founder of GPU Users Group |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
When I tested it in Linux all I had to do was use the Non-SoG version to solve the Best Gaussian problem. If you look at the Apps page it shows Windows as the only platform that doesn't have a Non-SoG nVidia App. I would suggest you build a Non-SoG Windows App and see if it has the Gaussian problem, I'd wager it won't. To use another app doesn't mean to fix the bug. Any bug in any software could be "fixed" by abandon app usage ;)
I'll take detailed look, thanks. SETI apps news We're not gonna fight them. We're gonna transcend them. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
version GLIBCXX_3.4.21 not defined in file libstdc++.so.6 with link time referenceThat's similar to the problems I saw with Ubuntu 14.04.1. You probably need a newer version of Mint with the static App. I've been able to compile other versions in Ubuntu 14.04.1 and even 12.04.5 that work with the older systems, but, I'm still testing it. For now you'll probably need Mint 18.2 for Petri's App. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
So far the version of zi3xs2 compiled in Ubuntu 12.04.5 seems to be working normally. It still has the same occasional Bad Best Pulse it's had Forever, but, the biggest problem is all the Inconclusive Instant Overflows which account for over half the inconclusive count. Ever so often one of the Instant Overflows is given an Invalid. If those Overflow Inconclusives could be reduced it would look Much better. I haven't been able to compile it in anything other than CUDA 9, so, if the CUDA 9 version doesn't work on the Kepler cc 3.5 GPUs there isn't much hope for them. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
@TBar Lets continue discussion on required Mac ATi apps modification on beta site. I posted some questions there please respond (there). SETI apps news We're not gonna fight them. We're gonna transcend them. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Ever so often one of the Instant Overflows is given an Invalid. Does wingman report overflow (different set but overflow) also? SETI apps news We're not gonna fight them. We're gonna transcend them. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
This is a Typical Invalid Overflow, https://setiathome.berkeley.edu/workunit.php?wuid=2708379644 The way I remember it, the CUDA App looks for Triplets First. If a Task starts with many Triplets the Overflow result will be 30 Triplets. If the App looks for Something Else First, such as the SoG App, then the results will most likely be 30 of whatever it is looking for, i.e. Not Triplets. |
Jeff Buck Send message Joined: 11 Feb 00 Posts: 1441 Credit: 148,764,870 RAC: 0 |
This is a Typical Invalid Overflow, https://setiathome.berkeley.edu/workunit.php?wuid=2708379644It actually may not be that simple. During an email exchange I had with Petri about a month ago regarding this problem, he said "I looked at my code and the pulses are checked before triplets. So it is not so an easy fix I thought. I will have to debug why my code misses many pulses on noisy packets and then some on 'normal' data." To complicate it further, it seems to be a problem that already exists in the older Cuda apps, as I noted previously, so it may be in some code that Petri's app actually inherited from the stock Cuda code. It just never surfaced until the 4-bit WUs started to flow. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
I'm seeing a different type of Invalid Overflow now...and it looks Strange. There are 3 results, https://setiathome.berkeley.edu/workunit.php?wuid=2708604299 SSSE3xj OS X 64bit Build 3551 SSE3xj Win32 Build 3584 Invalid x41p_zi3xs2, Cuda 9.00 special I'm still looking at these... |
petri33 Send message Joined: 6 Jun 02 Posts: 1668 Credit: 623,086,772 RAC: 156 |
This is a Typical Invalid Overflow, https://setiathome.berkeley.edu/workunit.php?wuid=2708379644It actually may not be that simple. During an email exchange I had with Petri about a month ago regarding this problem, he said "I looked at my code and the pulses are checked before triplets. So it is not so an easy fix I thought. I will have to debug why my code misses many pulses on noisy packets and then some on 'normal' data." Thank you Jeff, There is going to be a major overhaul of the code if the 'old' cuda code base is going ever to be compatible with SoG or any other version regarding to the pulses on overflow packets. The noise would induce an unnecessary lot of work to code in a parallel environment and induce a slowdown for what? An eye candy! !! ! !!!! ! !!!!! !!!!!!!!! !! !!!!!! ... An overflow is an overflow and any two similar apps can confirm that. If the apps are not the same then it will be asked from an another one if that it is a bad packet indeed. If nonagreement then ask anoher. etc. A 30/30 or a near miss is a noise bomb. BOOM! No Signature Here. EDIT: I'm a week off here starting now! |
©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.