Message boards :
Number crunching :
Linux CUDA 'Special' App finally available, featuring Low CPU use
Message board moderation
Previous · 1 . . . 60 · 61 · 62 · 63 · 64 · 65 · 66 . . . 83 · Next
Author | Message |
---|---|
Jeff Buck Send message Joined: 11 Feb 00 Posts: 1441 Credit: 148,764,870 RAC: 0 |
My comment wasn't intended to suggest that the problem didn't need to be fixed. I was just noting that I've seen it appear with the stock Cuda apps, in particular the Cuda50 running on my daily driver. And I was passing along your analysis that it wasn't just a processing order issue. Ideally, Jason would probably be the one to try to track it down in the current Cuda codebase, but he has been absent for awhile, so if it can be fixed in the Special App, I would expect that it could be ported back to the more widely used Cuda apps. As I think I've expressed multiple times previously, just because a WU overflows doesn't mean that it's worthless. That 30 signal cutoff was based on storage considerations, not the value of the scientific data. The apps need to report consistently and let the scientists sort through the results and make any "noise bomb" determination. Anyway, what appears to be a noise bomb to one person might actually turn out to be an alien ABBA concert to another. ;^) |
petri33 Send message Joined: 6 Jun 02 Posts: 1668 Credit: 623,086,772 RAC: 156 |
My comment wasn't intended to suggest that the problem didn't need to be fixed. I was just noting that I've seen it appear with the stock Cuda apps, in particular the Cuda50 running on my daily driver. And I was passing along your analysis that it wasn't just a processing order issue. ABBA! Yes. One night in a "pulse nightmare" (an old one "jungle") no. Choose your game. There are rules. Some hands in a Poker game give you points, some Ace in a sleeve does not. Change the rules - lower the limit to regard a 'signal' that is below the noise lever (i.e. below one scorewise) to be accepted. I could report all of them! To find the best that is not reported is time consuming in a parallel world. Be prepared for the quantum era. Mamma Mia! No, no. The Winner does not take it all - it is just on the wings of the next hurricane named after a man - Fernando. - ABBA quote off. Still no name until I fall asleep. 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 |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
To find the best that is not reported is time consuming in a parallel world. Yes. indeed. But still it's part of algorithm. Until algorithm will be changed best should be found correctly (some sort of reduction from best per CU to single best could be used to reduce slowdown from serialization. Regarding overflows - yep, early versions of SoG had same issue too. The more distributed task computation is the bigger amount of signals one should store to properly reorder on reporting. At some point it will too costly indeed. But if you still doing one icfft per kernel call amount of signals to keep should be not too huge. SETI apps news We're not gonna fight them. We're gonna transcend them. |
petri33 Send message Joined: 6 Jun 02 Posts: 1668 Credit: 623,086,772 RAC: 156 |
To find the best that is not reported is time consuming in a parallel world. +1, from the far end of the visible world. I'll be back. 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 |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
I'll be back.I'm starting to accumulate quite a few Invalid Overflows on the 750Ti machine. All three of my machines are running the same version of zi3xs2 compiled in Ubuntu 12.04.5, don't know why that machine is getting more Invalids. This last one included a machine running the old CUDA 4.2 App, and it is a bit different than zi3xs2; 24fe07ad.6233.23794.15.42.0 x41p_zi3xs2, Cuda 9.00 special Find triplets Cuda kernel encountered too many triplets, or bins above threshold, reprocessing this PoT on CPU... err = 1 x41p_zi3t1d, Cuda 8.00 special Find triplets Cuda kernel encountered too many triplets, or bins above threshold, reprocessing this PoT on CPU... err = 1 x41zi (baseline v8), Cuda 4.20 Find triplets Cuda kernel encountered too many triplets, or bins above threshold, reprocessing this PoT on CPU... |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
. . Hi Laurent, . . I have not given any feedback on your revision of your app. It is working very very well and I am having great success with it. It is very fast in terminal window. Stephen :) |
Bruce Send message Joined: 15 Mar 02 Posts: 123 Credit: 124,955,234 RAC: 11 |
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. Hi TBar, I would be willing to try the new s2 version. How would I need to change things? Will I need the Cuda 9 libraries? Newer driver not a problem to install. Any help appreciated. If the new cuda apps are not going to work with older hardware, we need to know. Thanks. Bruce |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
Back in Petri's original post he stated. This executable is only for those who can test offline on L I N U X and GTX10x0 !!! sm_61 i.e. GTX10x0 only !!!.. You don't need to download the CUDA 9.0 libraries as the executable has them statically linked internally. You would also need the latest drivers with CUDA 9.0 support. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Hi TBar,Check your PMs and let me know how that works. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
If anyone else has a Kepler cc 3.5 GPU post a link to the machine and I'll give you a link to the zi3xs2-35 App. I really don't see any difference with this App verses the ones compiled with the cc 5.0 & 5.2 flags on my machines. The first time you run it on a different GPU it will take about 20 seconds to compile and cache the correct GPU code, after that first time the run times are about the same as an App compiled with the different sm flags. It should work on Kepler cc 3.5 and higher GPUs in Ubuntu 12.04.5 and higher with driver 384.xx. The objective is to see if CUDA 9 will work on the cc 3.5 GPUs. |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
If anyone else has a Kepler cc 3.5 GPU post a link to the machine and I'll give you a link to the zi3xs2-35 App. I really don't see any difference with this App verses the ones compiled with the cc 5.0 & 5.2 flags on my machines. The first time you run it on a different GPU it will take about 20 seconds to compile and cache the correct GPU code, after that first time the run times are about the same as an App compiled with the different sm flags. It should work on Kepler cc 3.5 and higher GPUs in Ubuntu 12.04.5 and higher with driver 384.xx. The objective is to see if CUDA 9 will work on the cc 3.5 GPUs. . . Hi TBar . . OK, here's the thing ... I do have a Kepler CC3.5 card (GT730) but the only spare slot I can put it into is the i5-6600 with the 2 x 970s. . . So I don't think that would be a viable setup. The 970s were running fine with a GTX1050 in that slot until the weather got too warm but I don't think that a GT730 would play well with the 970s. If you feel otherwise let me know and I will give it a try. . . The only other alternative is to remove the 1050ti from the Core2 Duo and put it in there. That might be a viable test but would the S2 version be OK when I return to the 1050ti?? Stephen ?? |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
...but would the S2 version be OK when I return to the 1050ti??If it wasn't, what would stop you from going back to what you are using now??? Have you ever run the 730 with the Existing CUDA 8 & 6.5 Apps? If you decide to try the newer App, it would be nice to be able to compare it to the Existing Apps. |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
Are we still in pre-production testing for the CUDA 9.0 apps? Or are they usable in production now? Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
W3Perl Send message Joined: 29 Apr 99 Posts: 251 Credit: 3,696,783,867 RAC: 12,606 |
Thanks for the report :) Glad to know it could be useful to you. I have updated the script once again to fix some issues, you can retrieve it here |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
...but would the S2 version be OK when I return to the 1050ti??If it wasn't, what would stop you from going back to what you are using now??? . . Sadly I replaced the 730 with the 1050ti before the general release of the CUDA6.5/8.0. I can give you comparisons with SoG but not with earlier versions of the special sauce. . . I probably still have the downloads of an earlier version on that machine if you need a comparison base. . . And you didn't give a response to the question of which setup you think I should try. Though it is my impression it is the second option that you think is most viable. Stephen ?? |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
...but would the S2 version be OK when I return to the 1050ti??If it wasn't, what would stop you from going back to what you are using now??? . . I have looked on this machine and I still have downloads for 3k+ and 3t2b. Stephen ?? |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
I would recommend the Machine with the single GPU. It's usually best to test with just one GPU, and if you wish you could place the 1050Ti in the Multi-GPU machine. Use the current CUDA Apps, you can start with what's already installed. It should be a simple matter of just suspending all the GPU tasks in BOINC and then shutdown the machine and swap the cards. Once running just resume One GPU task and see how it works with the 730. If it works, run a few more and then Swap the CUDA Apps. I think someone said the CUDA 6.5 App is demanding the CUDA 6.5 Libraries instead of the 6.0 Libraries in the Links. I'll see if I can Upload the 6.5 Libraries to C.A. so you don't have to download the 6.5 ToolKit. To change the Apps just paste the App & Libraries into the setiathome.berkeley.edu folder and change the app_info.xml to name the New App & Libraries. The Library names don't matter to the CUDA 9 App, it will just use the Built-in Libraries anyway. The Link is Up for the 6.5 Libraries at C.A. http://www.arkayn.us/forum/index.php?topic=197.msg4499#msg4499 |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
I would recommend the Machine with the single GPU. It's usually best to test with just one GPU, and if you wish you could place the 1050Ti in the Multi-GPU machine. Use the current CUDA Apps, you can start with what's already installed. It should be a simple matter of just suspending all the GPU tasks in BOINC and then shutdown the machine and swap the cards. Once running just resume One GPU task and see how it works with the 730. If it works, run a few more and then Swap the CUDA Apps. I think someone said the CUDA 6.5 App is demanding the CUDA 6.5 Libraries instead of the 6.0 Libraries in the Links. I'll see if I can Upload the 6.5 Libraries to C.A. so you don't have to download the 6.5 ToolKit. To change the Apps just paste the App & Libraries into the setiathome.berkeley.edu folder and change the app_info.xml to name the New App & Libraries. The Library names don't matter to the CUDA 9 App, it will just use the Built-in Libraries anyway. . . OK, that would save time, but I was in the process of running down the cached tasks. I am running 3v at the moment so is there any issue with that and Kepler cc3.5 cards? I seem to remember people having issues with GTX780 units. . . And that is a good idea of putting the 1050ti into the i5 rig. It is low profile and should not restrict airflow as much as the 1050 I had in there before. Stephen ?? |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
Are we still in pre-production testing for the CUDA 9.0 apps? Or are they usable in production now? [bump] Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
. . OK, that would save time, but I was in the process of running down the cached tasks. I am running 3v at the moment so is there any issue with that and Kepler cc3.5 cards? I seem to remember people having issues with GTX780 units.This is what you call Testing. That is why I suggested suspending All your remaining GPU tasks before swapping cards. That way you can Test them One at a Time. There isn't any need to run down the cache as long as you DON'T Change the Version Number or Plan Class in your app_info.xml. Use the Same app_info.xml, all you need to change is the two lines where the App is named and the 4 lines where the Libraries are named...Nothing Else. No one has reported if the 730 works with zi3v or not, so, you are going to Test it. If it doesn't work, suspend the One task again, stop BOINC, and change Apps and Libraries. Of course, you should make sure BOINC sees the card before resuming the task. You will have to update the Driver to test the CUDA 9 App, I suggest this driver, http://www.nvidia.com/Download/driverResults.aspx/123918/en-us While you're at it, I would suggest updating to BOINC 7.8.2 so your driver version will be listed. Absolutely nothing wrong with 7.8.2, it works great. |
©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.