Message boards :
News :
New app v1.4 with OpenCL support deployed
Message board moderation
Author | Message |
---|---|
Send message Joined: 20 Apr 11 Posts: 388 Credit: 822,356,221 RAC: 0 |
Finally we have a newer application version that supports OpenCL! Updated versions of both AMD/ATI Stream and OpenCL apps for Windows have been offered by our server since yesterday to participants meeting requirements. These apps require BOINC Client v7.2.28 or newer and Windows XP or newer. The Stream app is not available to AMD Radeon HD 7870/7950/7970/R9 280X series (Tahiti) or newer cards due to compatibility issues. OpenCL app is preferred for those cards. The old app version is still available for systems running older BOINC Client. However, if your system has multiple GPUs, please consider updating it's BOINC Client to a support version. There are known issues running the old app on systems with multiple GPUs. New apps fix these by running one cruncher per device. For detailed app change history, please read application changelog. For latest requirements enforced by our server, please read latest requirements. Please report any problems you might encounter with new apps and our changes. Especially if you no longer get any work from our server where you previously did. You can report them by posting at our forum. Happy crunching! |
Send message Joined: 15 May 11 Posts: 1392 Credit: 5,774,953 RAC: 0 |
Tested. http://moowrap.net/result.php?resultid=29697972 Working good. |
Send message Joined: 14 Jul 11 Posts: 5 Credit: 14,457,478 RAC: 0 |
Tested. Working up to now VERY good. http://moowrap.net/show_host_detail.php?hostid=15107 First time i ever get GPU moo's. Nothing "tweaked". AMD 7670 has 1700 to 2000 sec runtimes. AMD 7560D aka AMD's APU has ~3000 sec runtimes. |
Send message Joined: 20 Jul 13 Posts: 5 Credit: 54,854,826 RAC: 93 |
Will Try Today Thank You ! EDIT : It's Working ! THANK YOU FOR YOUR HARD WORK !!!!! |
Send message Joined: 17 May 11 Posts: 3 Credit: 409,251,638 RAC: 0 |
ALL Jobs fail after a short runtime of 20 - 50 seconds on a 6470M. 9/7/2014 2:44:24 PM | | CAL: ATI GPU 0: AMD Radeon HD 6350/6450/7450/7470 series (Caicos) (CAL version 1.4.1703, 1024MB, 991MB available, 480 GFLOPS peak) 9/7/2014 2:44:24 PM | | OpenCL: AMD/ATI GPU 0: AMD Radeon HD 6350/6450/7450/7470 series (Caicos) (driver version 1084.4 (VM), device version OpenCL 1.2 AMD-APP (1084.4), 1024MB, 991MB available, 480 GFLOPS peak) 9/7/2014 2:44:24 PM | | OpenCL CPU: Intel(R) Core(TM) i7-2640M CPU @ 2.80GHz (OpenCL driver vendor: Advanced Micro Devices, Inc., driver version 1084.4 (sse2,avx), device version OpenCL 1.2 AMD-APP (1084.4)) The Parameters in the line "LLVM ERROR: Cannot select:" are different in every Job (eg.: http://moowrap.net/result.php?resultid=29696345): Stderr Ausgabe <core_client_version>7.4.12</core_client_version> <![CDATA[ <stderr_txt> 15:41:36 (13620): wrapper v1.4 build 16 for AMD/ATI OpenCL starting (BOINC Wrapper v7.5.26011) 15:41:36 (13620): device: OpenCL: AMD/ATI GPU 0: AMD Radeon HD 6350/6450/7450/7470 series (Caicos) (driver version 1084.4 (VM), device version OpenCL 1.2 AMD-APP (1084.4), 1024MB, 991MB available, 480 GFLOPS peak) 15:41:36 (13620): checkpoint interval: 0h15m00s00 (task 56000 GFLOPS, 0h01m56s66 per packet) 15:41:36 (13620): wrapper: running dnetc520-win32-x86-opencl.exe (-ini dnetc.ini -runoffline -multiok=1) - attempt 1/10 LLVM ERROR: Cannot select: 0x3da1b68: i8 = setcc 0x3d9fe00, 0x3da2140, 0x3da1a58 [ID=41] |
Send message Joined: 14 Jul 11 Posts: 5 Credit: 14,457,478 RAC: 0 |
As far as I can read in AMDs dev forum and here this seems to be a result of a driver problem. some driver versions cannot use specific code inside opencl. You may consider a driver update. |
Send message Joined: 4 May 11 Posts: 27 Credit: 112,091,698 RAC: 0 |
It's doing fine on my R9 290X ! :) I had to set up CPU affinities manually to achieve best efficiency, like I hade to do with all OpenCL applications ... |
Send message Joined: 3 May 11 Posts: 1 Credit: 102,660,825 RAC: 0 |
I would like to choose between ATI stream and OpenCL, but I can't see any way to do it. W7 64bit, ATI R9 280, driver: 1.4.1848 OpenCL: 1.2, CPU X6 1090T, Boinc core 7.2.42. OpenCL application makes 67 % GPU load only, rising a priority of the CPU process has no effect. Run times are 1,3x longer compared to 1.03 ATI stream application. App_info.xml doesn't work, disabled on the server side? |
Send message Joined: 16 Dec 13 Posts: 24 Credit: 50,444,089 RAC: 79,172 |
So far, I've got one good unit in from my R9 280X and am working on a second... Awesome! This is a fairly new machine to me, and it's the first GPU project that really seems to be taxing my GPU, as the fans kick in to overdrive as compared to Einstein, SETI, etc. I used to use the Win 7 widget "GPU meter" from Addagadget.com on my old HD4670, though none of their versions seem to work with the new card so I don't know what the utilization is like. The old HD4670 would hover around 90% +/- 5% which was great for me because it didn't produce any UI lag like GPU work from other projects did. I haven't experience any UI lag from any projects with the new card, but I would like to know the utilization--does anyone else have a widget/utility working with their R9 280X? Mine is a SAPPHIRE 100363L Radeon R9 280X 3GB. Click here to see My Detailed BOINC Stats |
Send message Joined: 20 Apr 11 Posts: 388 Credit: 822,356,221 RAC: 0 |
...and it's the first GPU project that really seems to be taxing my GPU, as the fans kick in to overdrive as compared to Einstein, SETI, etc. That's a good sign for the GPU load. :)
I use GPU-Z to see what my GPU utilization is and usually when GPUs are crunching away they hover around 97% for both of my cards. Normally I only watch that my temps are highish so that at least something is crunching. :) -w |
Send message Joined: 20 Apr 11 Posts: 388 Credit: 822,356,221 RAC: 0 |
I would like to choose between ATI stream and OpenCL, but I can't see any way to do it. W7 64bit, ATI R9 280, driver: 1.4.1848 OpenCL: 1.2, CPU X6 1090T, Boinc core 7.2.42. OpenCL application makes 67 % GPU load only, rising a priority of the CPU process has no effect. Run times are 1,3x longer compared to 1.03 ATI stream application. If the run times are constantly slower, the BOINC Server should start favoring the other app for you after a while. At the moment there's no setting to force using only specific app but I might add it in the future. It's possible you need to play with core settings (they are in project preferences) to select more optimal one to tax the GPU more. You can run Dnet Client from the project dir with "dnetc520-win32-x86-opencl.exe -l \output.log -bench rc572" command and read through the log file to see what it recommends for you. I should somehow automate grabbing that benchmark or preferably the whole thing. App_info.xml doesn't work, disabled on the server side? The app_info.xml is fully on the client side, server doesn't have anything to say for it so it's not disabled. If you only have ATI Stream (grab the new versions, though) in it, you should only get jobs for that. -w |
Send message Joined: 22 Jun 11 Posts: 2080 Credit: 1,844,350,984 RAC: 71,156 |
I would like to choose between ATI stream and OpenCL, but I can't see any way to do it. W7 64bit, ATI R9 280, driver: 1.4.1848 OpenCL: 1.2, CPU X6 1090T, Boinc core 7.2.42. OpenCL application makes 67 % GPU load only, rising a priority of the CPU process has no effect. Run times are 1,3x longer compared to 1.03 ATI stream application. I saw this at another project just this morning, it applies to you too: "You need to run the OpenCL app on that machine. It is currently downloading the ATI app which requires CAL and your GPU is known to return invalid results when running CAL because AMD changed the instruction set with the HD 7xxx and later GPUs." This message, also from theo ther project may help explain another part of your problem: "It used to be that using a large lookup table in texture memory on the GPU worked best. It still does with nVidia hardware. With OpenCL, that no longer seems to be the case with AMD hardware. By reducing the lookup table to the point where it fits within AMD's cache, it can keep the stream processors busy. At present, the bittleneck for AMD GPUs is memory speed rather than processing speed. Sosiris has been doing quite a bit of R&D and has suggested a number of changes that improve performance on newer AMD GPUs." He is working to solve it on his project, Collatz, if he does I am sure the other projects will pick it up too. |
Send message Joined: 20 Apr 11 Posts: 388 Credit: 822,356,221 RAC: 0 |
The Parameters in the line "LLVM ERROR: Cannot select:" are different in every Job (eg.: http://moowrap.net/result.php?resultid=29696345): Hmm, according to http://devgurus.amd.com/thread/160579 that seems to be a bug in the driver. According to your driver version 1084.4 you are using Catalyst 13.1 drivers, is that correct? They claim the bug is fixed in Catalyst 13.4 or newer. Can you try upgrading your driver and see what happens? -w |
Send message Joined: 16 Dec 13 Posts: 24 Credit: 50,444,089 RAC: 79,172 |
I use GPU-Z to see what my GPU utilization is and usually when GPUs are crunching away they hover around 97% for both of my cards. Normally I only watch that my temps are highish so that at least something is crunching. :) Ah yes, GPU-Z seems to work, though I'd like to have a functional Widget so I can monitor things from the desktop. Moo (ati14) seems to hold the GPU at 99% with short drops to around 80% so I would say you nailed the Tahiti support with that one. I've only had one opencl_ati_101 unit thus far, it ran successfully with run time 3rd of 4 total new units (the others being ati14). In comparison, Einstein has a very low usage, with only short spikes of 30-80%. Bitcoin Utopia holds around 60%. This is with all 4 CPU cores crunching, if I pause the CPU load then utilization goes up. Good to see this project is making good utilization of my shiny new rig, though I may not be so happy when the utility bill comes. Now, if I could only get some running on my OS X machines (one CUDA, one ATI)... Click here to see My Detailed BOINC Stats |
Send message Joined: 22 Jun 11 Posts: 2080 Credit: 1,844,350,984 RAC: 71,156 |
Try 'gpuobserver.zip' it is a 'gadget' for your Windows desktop that shows stuff about your gpu. |
Send message Joined: 16 Dec 13 Posts: 24 Credit: 50,444,089 RAC: 79,172 |
Try 'gpuobserver.zip' it is a 'gadget' for your Windows desktop that shows stuff about your gpu. All it shows for my card is temperature. Looks like it's a pretty old widget, the last post on the developer's site is February 2011. I think that's the problem with my other one as well, it was last updated in January 2013. Click here to see My Detailed BOINC Stats |
Send message Joined: 4 May 11 Posts: 27 Credit: 112,091,698 RAC: 0 |
Tip for (any) OpenCL app on ATI cards : the GPU usage is usually maxed when the application runs with fixed CPU affinity setting. You can do this manually with Windows Task Manager by checking only one thread in affinity setting. If you see performance improvement, this trick can be automanted with Bill2's Process Manager which allows the création of rules to automatically apply settings when a given process is started. |
Send message Joined: 10 May 11 Posts: 17 Credit: 7,327,486,541 RAC: 20,906,864 |
I tested the new app on different platforms and everything went well. This new version solved a lot of annoying issues, you did a good job! Two more things will be useful: - A way to select between opencl and ati/cal/stream (stream is much better for HD6xxxx cards) - A way to choose the right core for opencl applications in the preferences pages. For instance the default (auto) selected core (core #1 CL 1-pipe) is slower than core #3 (according to benchmark results) on my 290-X. |
Send message Joined: 20 Apr 11 Posts: 388 Credit: 822,356,221 RAC: 0 |
I tested the new app on different platforms and everything went well. That's good to hear! - A way to select between opencl and ati/cal/stream (stream is much better for HD6xxxx cards) The way BOINC Server works is that it sends both variants and if one of them is constantly faster, it should start preferring that variant. It will still occasionally send the other one to see if it has gotten any better. I might add a project profile setting that allows one to disable OpenCL or native variant. - A way to choose the right core for opencl applications in the preferences pages. For instance the default (auto) selected core (core #1 CL 1-pipe) is slower than core #3 (according to benchmark results) on my 290-X. At the moment the AMD/ATI OpenCL app uses the ATI Stream setting (just as nVidia OpenCL app will use nVidia CUDA setting) to override the detected core for the app. Do you need a separate override for each type or is it enough for now? I might add two OpenCL specific core overrides for the next app version. -w |
Send message Joined: 16 Dec 13 Posts: 24 Credit: 50,444,089 RAC: 79,172 |
Try 'gpuobserver.zip' it is a 'gadget' for your Windows desktop that shows stuff about your gpu. The update to Catalyst 14.9 seems to have "fixed" the widget, now it shows temperature, fan speed, GPU load, clock speed and voltage. Click here to see My Detailed BOINC Stats |