Message boards :
News :
New GPU OpenCL versions available
Message board moderation
Author | Message |
---|---|
Send message Joined: 8 Jul 11 Posts: 1341 Credit: 509,743,096 RAC: 555,135 |
I just deployed the windows AMD openCL version as a beta app. I will monitor results for errors, but please report any strange behavior. I also deployed newer openCL versions for AMD linux and Nvidia windows. The openCL code had a couple minor tweaks; nothing major. |
Send message Joined: 16 Feb 17 Posts: 2 Credit: 671,890 RAC: 0 |
Dear administrator I tried to get some opencl workunits on Linux amd64 with RX 560 OpenCL, but the scheduler does not send me any. On the applications page, I see there is app available for my platform. I tried with and without beta enabled, but it makes no difference. Sending scheduler request: Requested by user. Requesting new tasks for CPU and AMD/ATI GPU Scheduler request completed: got 0 new tasks Tasks for CPU are available, but your preferences are set to not accept them |
Send message Joined: 28 Oct 11 Posts: 180 Credit: 246,438,014 RAC: 161,179 |
You may find you need to check the matching box for the right venue on https://numberfields.asu.edu/NumberFields/prefs.php?subset=project and update twice - once to update the client settings, and again to actually use the new ones. |
Send message Joined: 8 Jul 11 Posts: 1341 Credit: 509,743,096 RAC: 555,135 |
Hi Tomas, I think the reason is because the plan class requires openCL 1.2 or higher, but it looks like your GPU is configured to use 1.1. Would it be possible to upgrade your GPU driver to use 1.2? |
Send message Joined: 21 Mar 17 Posts: 3 Credit: 605,360 RAC: 334 |
Hi, I try OpenCL_AMD , workunit starts to run after 3-4 minutes stops and says "waiting for memory". I try suspend-resume, Close boinc and restart, waiting 15+minutes nothing happens just waiting. Abort the WU and get another one same problem at %75 it starts waiting memory, also When I look GPU no used cpu full use one core at first %75. Task properties as follows Application Get Decic Fields 3.05 (opencl_amd) Name wu_sf3_DS-13x271_Grp1576298of6553600 State Waiting for memory Received 04/11/2019 21:50:50 Report deadline 11/11/2019 21:50:50 Resources 0.879 CPUs + 1 AMD/ATI GPU Estimated computation size 7,000 GFLOPs CPU time 00:10:13 CPU time since checkpoint 00:10:13 Elapsed time 00:10:16 Estimated time remaining 00:03:17 Fraction done 75.707% Virtual memory size 11.68 GB Working set size 9.59 GB Directory slots/0 Process ID 8940 Progress rate 7.374% per minute Executable GetDecics_3.05_windows_x86_64__opencl_amd |
Send message Joined: 8 Jul 11 Posts: 1341 Credit: 509,743,096 RAC: 555,135 |
Hi, I try OpenCL_AMD , workunit starts to run after 3-4 minutes stops and says "waiting for memory". I try suspend-resume, Close boinc and restart, waiting 15+minutes nothing happens just waiting. Abort the WU and get another one same problem at %75 it starts waiting memory, also When I look GPU no used cpu full use one core at first %75. So during testing we learned that older GPUs have problems with the NF app, and I think that's what's happening here. We noticed cards made before ~2017 have problems and it looks like your card is the R9 200 series which was made in 2013. Sorry about that. |
Send message Joined: 12 Oct 13 Posts: 17 Credit: 39,780,803 RAC: 4,996 |
Hi Eric, Does AMD or Nvidia offer better performance? How would you rate the performance between your 3 GPUs: GTX 1660, GTX 1050, and the unidentified AMD Ellesmere [Radeon RX 470/480/570/570X/580/580X/590] Is there any difference in CPU utilization between GPU apps? It appears the Turning based Nvidia cards have good performance, so I have been looking at a 1660 series card to replace an aging AMD 7970. Thanks, Matt |
Send message Joined: 8 Jul 11 Posts: 1341 Credit: 509,743,096 RAC: 555,135 |
Hi Eric, I only got interested in GPU computing about a year ago, so I am no means an expert. But I'll give you my opinion. I personally prefer the Nvidia cards. I think AMD is focused more on the gaming industry as opposed to scientific computing. AMD only officially supports windows and a couple linux distributions. It took me many hours to finally get my Radeon to actually work and it's still about 30% slower than the GTX 1050 when it should be a little faster. To answer your question regarding performance, at least on this project, I believe the Nvidia cards are better. My GTX1660 is kickin butt (12x faster than a cpu core). Then comes the GTX1050 which is about 5x faster than a cpu core). Finally, the RX570 is about 4 times faster than a core. Another way to look at it, here are the average run times from the 13x271 batch: GTX1660: 5.0 min GTX1050: 10.5 min RX570: 14 min There should be no difference in cpu utilization between the app versions. Hope that answers all your questions. |
Send message Joined: 12 Oct 13 Posts: 17 Credit: 39,780,803 RAC: 4,996 |
Thank you. That's just what I was looking for. |
Send message Joined: 14 Mar 19 Posts: 9 Credit: 105,478,826 RAC: 89,321 |
Wise to have a stable AMD GPU computing program as a RX 5700 user. lol |
Send message Joined: 21 Sep 19 Posts: 1 Credit: 3,183,711 RAC: 0 |
Hi Eric, I hope you enjoyed your holiday. Possible to make GPU CUDA on a Mac work ? best regards Brice |
Send message Joined: 8 Jul 11 Posts: 1341 Credit: 509,743,096 RAC: 555,135 |
Hi Eric, Actually, I'm still on holiday for a few more days. Off hand I'm not sure how I can make cuda work for the Mac, but I will think about it. |
Send message Joined: 13 Sep 11 Posts: 3 Credit: 6,843,264 RAC: 0 |
Today I tried to upgrade to nvidia driver 440 which is the latest... Sorry to say but when installed the newest driver my boincmanager(7.9.3) that comes with ubuntu 18.04 in programs to install. The gpu is missing. And that is happening on both my computors. So now after 4 hour of looking for what is wrong i returned to 410.93 drivers which have gpu(ocl) support for boinc manager. 1 computor that is... The second computor lost all network both wireless and ethernet... I guess that comp needs a re-install now.... Nvidia latest 440 does not werk properly with boinc manager 7.9.3 |
Send message Joined: 8 Jul 11 Posts: 1341 Credit: 509,743,096 RAC: 555,135 |
Today I tried to upgrade to nvidia driver 440 which is the latest... That sucks. I use fedora and I have Nvidia driver 440.44 installed (this was the latest as of a few weeks ago). I use version 7.8.4 of the BOINC manager. So I am not sure exactly what caused your problem. I have had issues in the past if I didn't do a complete uninstall of the old driver before installing the new one. |
Send message Joined: 13 Sep 11 Posts: 3 Credit: 6,843,264 RAC: 0 |
The 440.44 driver from nvidia works.... download here: https://www.nvidia.com/en-us/drivers/unix/ I used some ppa stuff and it downloaded and so on but never worked properly. I successfully run the 440.44 on my boincmanager 7.9.3 and it detects gpu and runs app from numberfields...ubuntu 18.04 /TRuEQ |
Send message Joined: 13 Sep 11 Posts: 3 Credit: 6,843,264 RAC: 0 |
I added a second box now. All looks nice and gpu temps are very good. |
Send message Joined: 24 May 19 Posts: 38 Credit: 1,383,935 RAC: 1,022 |
hi everyone, I have a problem with this app. Passing the calculation hours, the percentage of progress remains at 9.996%. Also when I close boinc and then turn off the PC, the GPU remains 99% occupied, crashing either the whole PC or the AMD driver. Does it just happen to me? For more information on my PC here is the link: https://numberfields.asu.edu/NumberFields/show_host_detail.php?hostid=1612339 |
Send message Joined: 8 Jul 11 Posts: 1341 Credit: 509,743,096 RAC: 555,135 |
hi everyone, I have a problem with this app. The 9.996% point is usually the first time the data gets offloaded to the GPU for processing. If it hangs there, that tells me its the GPU that's crashed. Strange that rebooting does not help. After it reboots, could it be starting the boinc client automatically and putting you back into the same bad state? Try suspending the GPU and/or the project from the boinc manager before rebooting to see if that helps. |
Send message Joined: 24 May 19 Posts: 38 Credit: 1,383,935 RAC: 1,022 |
So I tried to suspend the wu, close the boinc client and restart the PC, then restart the boinc client and the WU and the progress percentage restarts from 0% and after 30 seconds it returns to 9.996% and then remains stuck there. I also tried suspending the wu and the percentage was always 9.996%, after which the client boinc downloaded me a wu of Seti@home beta x GPU opencl and before starting the processing of the wu, the AMD RADEON software restarted . In the meantime, I removed the suspension from the wu Numberfields@home, when, after having finished processing the wu from Seti, it always started from the pecial 9.996% and from there it never moved. This morning I restarted the PC and started the client boinc, the wu as always started from 0% and now after 14 hours of processing it is still stopped at 9.996%. I looked at the stderr.txt file in the wu processing slot folder and noticed these two entries: CHECKPOINT_FILE = wu_sf3_DS-14x271_Grp1326090of3932160_checkpoint. Checkpoint Flag = 0. Are they right? In fact there is no checkpoint file on my PC. Sorry for the length of the message, but I had to explain the various tests made. |
Send message Joined: 8 Jul 11 Posts: 1341 Credit: 509,743,096 RAC: 555,135 |
So I tried to suspend the wu, close the boinc client and restart the PC, then restart the boinc client and the WU and the progress percentage restarts from 0% and after 30 seconds it returns to 9.996% and then remains stuck there. It sounds like a problem with the AMD opencl driver on windows. I personally have not tested that particular combination. I do know the nvidia opencl driver on windows works and the amd opencl driver on linux also works, so I know the opencl code is good. I just queried the database for results from the amd opencl windows version and there were plenty of successful results. If one of those users happens to see this maybe they could let you know what driver version they are using, or if they had to do anything special to make it work. Other than that, it's probably not worth your time messing with it anymore (unless you enjoy that kind of thing). |