Message boards :
Number crunching :
Version 8.12?
Message board moderation
Author | Message |
---|---|
Mark Loukko Send message Joined: 7 Jun 99 Posts: 52 Credit: 40,406,567 RAC: 108 |
I just received a few work units with the following: V8 8.12 (opencl_nvidia_SoG) V8 8.12 (opencl_nvidia_sah) Normally my work units are called (cuda50) version 8.00 not 8.12. Anybody else getting these? Also, I'm not sure what SoG stands for. The estimated computation size is ~180,000 GLOPS (same as my normal cuda50) but the estimate crunch time is about an hour longer. I'm also seeing these for Green Bank work units. Cheers Mark |
Brent Norman Send message Joined: 1 Dec 99 Posts: 2786 Credit: 685,657,289 RAC: 835 |
It looks to me that those are just appearing on your Intel GPU. |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
No, you can find them on the applications page. Windows/x86 8.12 (opencl_nvidia_sah) 18 May 2016, 1:10:51 UTC Windows/x86 8.12 (opencl_nvidia_SoG) 18 May 2016, 1:10:51 UTC Not sure if its a new stock release or what... |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
OK, I've been waiting for the 8.12 Nvidia_SoG app to appear on Main. Is this the same version that has been running on Beta? Is the app that is on Mike's World website the same version as on Main now? I run the Anonymous platform. Is there going to be a new Lunatics installer that incorporates the new 8.12 app? If so... when? Can anyone point to the definitive thread and message that lists the recommended command line parameters? I believe I saw that you have to have both app_info AND app_config files in agreement to be able to run multiple tasks on each GPU in the system. I need to know the specific settings to accomplish this task. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
Keith, I would recommend just running 1 work unit at a time for now. See how your system responds. Yes, when you run more than 1 you need to have commandlines and app_config.xml in agreement on the number of instances. But if you only run 1 at a time, not an issue. Also, the app requires 1 core for 1 work unit. Again not a big deal with 1 work unit. Becomes an issue when 2-3 with a smaller CPU chip. Using -use_sleep can help reduce that high CPU usage but with older cards it (that command) can cause problems. Edit.. Keith hold off on anything just yet. Once I get back tomorrow I can help you more with specifics. |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
Zalster, thanks for the reply and offer of help. I would like to run two tasks per card if I can. I have two cards in each machine. I think that is what is going to trip me up. I think I have enough CPU cores to handle four tasks per machine since I already have been running only 4 CPU tasks and using the other 4 to support feeding the dual GPU's. I have been using the sleep command in the old V7-V8 Nvidia command files to allow me to use the main driver for the normal duties of a computer and that has worked well for me so far until the Guppi VLAR's showed up on the GPU's the last outrage. I am suffering some latency now in text entry and mouse cursor that is bothering me. The other computer is just a cruncher and any latency on it is never noticed. I will wait until someone posts some recommended parameters. I have been crunching Guppi VLAR's in just under an hour so far for the most part with the stock Lunatics app, but I believe I can get that down to about half an hour with the new SoG app I think based on what I have seen in the threads I've lurked in. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
Keith, Here is the link to Raistmer's apps You want the r3430_SoG version. You will need something to extract the files as it's a zip https://cloud.mail.ru/public/DMkN/x4BRCYuAV Move all of the component into you seti@home folder. You will need to remove the cuda section of your app_info and replace it with the one for the opencl <app> <name>setiathome_v8</name> </app> <file_info> <name>MB8_win_x86_SSE3_OpenCL_NV_r3430_SoG.exe</name> <executable/> </file_info> <file_info> <name>libfftw3f-3-3-4_x86.dll</name> <executable/> </file_info> <file_info> <name>mb_cmdline_win_x86_SSE3_OpenCL_NV.txt</name> </file_info> <app_version> <app_name>setiathome_v8</app_name> <version_num>800</version_num> <platform>windows_intelx86</platform> <avg_ncpus>0.04</avg_ncpus> <max_ncpus>0.2</max_ncpus> <plan_class>opencl_nvidia_SoG</plan_class> <cmdline></cmdline> <coproc> <type>CUDA</type> <count>0.5</count> </coproc> <file_ref> <file_name>MB8_win_x86_SSE3_OpenCL_NV_r3430_SoG.exe</file_name> <main_program/> </file_ref> <file_ref> <file_name>libfftw3f-3-3-4_x86.dll</file_name> </file_ref> <file_ref> <file_name>mb_cmdline_win_x86_SSE3_OpenCL_NV.txt</file_name> <open_name>mb_cmdline.txt</open_name> </file_ref> </app_version> If you want APs, just leave in the section for APs in your app_info.xml As stated previously, the app wants to use 1 CPU core for each work unit. To decrease this, use the -use_sleep command in the mb_cmdline.txt This command works fine with newer GPUs but has problems with older GPUs This is for my Titan Xs and 980Ti. You might have to decrease the period_iterations_num to 40 to start with -use_sleep -sbs 512 -total_GPU_instances_num 4 -instance_per_device 2 -period_iterations_num 80 -spike_fft_thresh 4096 -tune 1 64 1 4 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 64 -oclfft_tune_cw 64 -cpu_lock -hp This is 1 of 2 places that must match to get the correct number work unit on each GPU Total gpu instance num is the total across all gpus for you computer, so for you it's 4 instance per device is 2 per gpus Lastly, your app_config.xml must match the instances per device, ie when you set the ratio it should look like this. plan class is the OpenCL_nvidia_SoG <app_config> <app_version> <app_name>astropulse_v7</app_name> <plan_class>opencl_nvidia_100</plan_class> <avg_ncpus>0.35</avg_ncpus> <ngpus>0.5</ngpus> </app_version> <app_version> <app_name>setiathome_v8</app_name> <plan_class>Opencl_Nvidia_SoG</plan_class> <avg_ncpus>0.35</avg_ncpus> <ngpus>0.5</ngpus> </app_version> </app_config> I would recommend starting with 1 work unit per GPU so make the changes to both this commandline and app_config.xml so you start with 1 and work your way up. If you choose not to use the -use_sleep you will find the computer may get sluggish. So I don't recommend not using it. As you can see, I don't run CPU work units anymore since I am not doing them on the GPU. It helps to keep the system from getting to slow. Edit.. If you choose to try the standard stock version, you can still use the app_config to control the ratio and the commandline. I need to find where you need to put the extra _ in plan class, Raistmer was talking about Eric liking to place an extra one in the exe name so give me a few minutes to research Edit.. Ok, sorry it was Richard who make the comment. They were talking about this setiathome_8.12_windows_intelx86__opencl_nvidia_SoG.exe Which if you download and install directly from Raistmer's site doesn't matter since they don't use that name. But if you try to build your own app_info with stock you need to get the name just as it is listed from Seti. |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
Ok, I realized I wasn't as clear as I should have been The command lines go here mb_cmdline_win_x86_SSE3_OpenCL_NV.txt Open with notepad and paste in that line. Make sure there is a space before the first - |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
I never had one in front of mine as well but I can't remember what thread it was mentioned that the space at the beginning of the command line for the MB needed a space. |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
Thank you, thank you Zalster. This is the specifics I needed to get started on the new SoG app. What I didn't realize is that you need to remove the old CUDA app sections, that would have tripped me up for sure. I believe I have the necessary understanding of the various parameters used in the files now. Now just have to back up the directory and old app_info and app_config files and start editing. I hope that setting the manager for no internet access upon first try and startup will prevent me from ghosting my tasks if I screw up. I have never had to restore my BOINC directory before to get out of a screw up situation. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
I never had one in front of mine as well but I can't remember what thread it was mentioned that the space at the beginning of the command line for the MB needed a space. I have seen more than one person mention that, but I have never used a space at the beginning of my commandline.txt files. I see all of my specified commands reflected in the Stderr output. Perhaps there was a beta/test version that had an issue where that was once required & it has been incorrectly propagated since? I'm not sure if - is being use as a delimiter or if whole command switches are being read, but I would not expect a space before the first command to be relevant. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
"-" used as part of parameter name, delimeter is spacebar. |
Mark Loukko Send message Joined: 7 Jun 99 Posts: 52 Credit: 40,406,567 RAC: 108 |
I've run about 30 version 8.12 SoG work units so far. My observations: - Green Bank units are now completing in about 25 minutes. Before v8.12 they were taking 37 minutes. In both cases I was running a single WU on my GPU. - Arecibo times are unchanged at ~10 minutes. - The GPU load for a single WU is now quite high, around 95%. The GPU hovered around 65% for a single WU before v8.12 - The CPU seems to be quite a bit busier with 8.12 but I don't have any numbers to back it up. - The TDP for the GPU hovers around 80%. Before it was around %50. BOINC screen shot: https://drive.google.com/open?id=0B_nA7JjIKorXRzRBcUl6bkU1VWc GUP-Z screen shot: https://drive.google.com/open?id=0B_nA7JjIKorXYW1ZWHNmazlzNE0 Cheers Mark |
©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.