Questions and Answers :
GPU applications :
GPU dont do more than 50% Core Load !
Message board moderation
Author | Message |
---|---|
Stauning Send message Joined: 10 Nov 00 Posts: 25 Credit: 67,819,196 RAC: 749 |
My GTX1080 only get around 50% core load, even when the only thing running on the PC is Boinc with Seti. Is there anyway Ill can get this little PC to get a higher GPU load? Michael. |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
My GTX1080 only get around 50% core load, even when the only thing running on the PC is Boinc with Seti. Let me see if I understood the question. Your Video card is only loading the task manager to 50% of the available cpu resources? Are you running any Seti tasks on the cpu? Are you running multiple tasks on your gpu? Are you using the mb*sog.txt command line in the seti data directory? Have you installed TechPowerup Gpu-Z? (https://www.techpowerup.com/gpuz/) What kind of gpu load does it report? Is it pegged pretty much at the 98-99% level? Have you installed TThrottle (http://www.efmer.com/b/?q=tthrottle_download_version) ? HTH, Tom A proud member of the OFA (Old Farts Association). |
Stauning Send message Joined: 10 Nov 00 Posts: 25 Credit: 67,819,196 RAC: 749 |
Well my PC is useing both CPU and the GPU. CPU load is limited to 75% of the cores, but up to a 100% load. GPU has no limit. Avg CPU use the last 48 hours is: 97% with a short max of 99,8%. GPU load in that time, has a avg of 57% with a max of 96% very shortly. CPU/GPU loads are read with HWinfo. And looking at the PC's Gflops/s, then the GPU max at ~2.400gflop (Astropulse 100). at bit less than this PC got when it had a GTX780 ! Max temperture on the PC is never more than a delta 20c (some 40c) for the GPU, and a delta 40 on one of the CPU cores (60c). So something is wrong !!! And Ive done nothing else than install Boinc, and added Seti@home. Then changed the GFX, and rerun messure. Michael. |
OzzFan Send message Joined: 9 Apr 02 Posts: 15691 Credit: 84,761,841 RAC: 28 |
So something is wrong !!! Why do you assume that? What you're seeing is the default behavior of GPU tasks on SETI. |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
CPU/GPU loads are read with HWinfo. So what is the gpu load when checked with "Tech Power up" utility that I previously mentioned? HWInfo may be very well but since it is not the utility I am comparing your load with, it is an "apples/oranges" issue. Do you have your app_config.xml file set to use a core per task on the gpu? How many tasks are you running on your gpu? What are you using in the mb*sog.txt command line file? (or app info file, if that is where you have the command line parameters). We need more details. Tom A proud member of the OFA (Old Farts Association). |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
My GTX1080 only get around 50% core load, even when the only thing running on the PC is Boinc with Seti. You have no computer listed that has a 1080 in it... The only 2 computers you have both do not have GPUs... Where is this computer that you are referring to? Is it under another profile? Did you pull it out of the computer? |
Stauning Send message Joined: 10 Nov 00 Posts: 25 Credit: 67,819,196 RAC: 749 |
Well Ive assume that something is wrong when a program dont use all the resurses it have to its dispostal. That PC is not running for me at the monent its doing seti for a friend.. Helping him getting no2 on the team :) Then Ive pull all my PC's over to my self - and starting doing some real Seti on my acount. http://setiathome.berkeley.edu/show_host_detail.php?hostid=8216884 Load with techpowerups utility is: 57% in avg over 2 hours of looking at it, with a 91% top for 1-2sec. app_config.xml ??? Where and how? Task on my GPU, one Ive guess... cant find a mb*sog.txt on my pc. So one of my problems may be that Ive have no clue, how to setup this "new" version of Seti (Been useing the classic alot). Michael. |
Stauning Send message Joined: 10 Nov 00 Posts: 25 Credit: 67,819,196 RAC: 749 |
Oki did some more reading and the mb*sog.txt your are talking about could be this one: %appdata%\boinc\projects\setiathome.berkeley.edu\mb_cmdline-8.22_windows_intel__opencl_nvidia_SoG.txt Its a emty file ! BUT, Ill have now made a app_config, and are now running two Nvidia_SoG, and the 1080 are at 98~99% load ! Thanks all for guideing me in the right direction. |
Stauning Send message Joined: 10 Nov 00 Posts: 25 Credit: 67,819,196 RAC: 749 |
BTW, this is also mine: https://setiathome.berkeley.edu/show_host_detail.php?hostid=8217028 It has been running Seti a shorter time than my 4790K with the GTX1080, and its doing more work !, yes its only hours shorter - but its a slower CPU, and a slower GPU. Oh and the X990 is not running Seti for 3-4hours a day. That was why Ill did think: There is something wrong here... |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
etiathome.berkeley.edu\mb_cmdline-8.22_windows_intel__opencl_nvidia_SoG.txt Yes, it's empty until you place commandline into it to tell it what to do. Those values are determined by the GPU that are being used. They also will vary depending on how hard you want the GPU to work, if the computer will be used for anything other than crunching and if lagging screens bother you are not.... You can probably find a wide assortment over in Number Crunching where we discuss these things fairly regularly |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
Yes, it's empty until you place commandline into it to tell it what to do. Those values are determined by the GPU that are being used. They also will vary depending on how hard you want the GPU to work, if the computer will be used for anything other than crunching and if lagging screens bother you are not.... From the mb_cmdline-8.22_windows_intel__opencl_nvidia_SoG ???readme ??? .txt file you will find the following for "high end" cards. It might speed up your gpu processing some. High end cards x8x 780TI Titan / Titan Z (and maybe yours) -sbs 256 -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 -hp -cpu_lock -high_perf (*requires testing) I would also add a "-hp" parameter at the end or beginning of the string. Don't include the (*requires testing) or the title above the line with the dashes. If you are willing to possibly tolerate a very slow responding screen you can try: -high_perf Another thing you could try is the -cpu_lock which might either speed up or slow down the processing. (According to the docs). The arbiter seems to be how fast the WU's are pushed through X number of processes/tasks running. HTH, Tom A proud member of the OFA (Old Farts Association). |
Stauning Send message Joined: 10 Nov 00 Posts: 25 Credit: 67,819,196 RAC: 749 |
So all Ive have to do, is copy that line and paste it in to mb*SoG.xml? And a slow screen dont bother me, its running Seti for 22-23 hours a day - and Iracing the rest of the time (with Boinc shutdown). Ive moved to PC over to my ovn account, since that will make it easier to fiddel with it :) Thanks. |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
Yes, copy it into that file, which is now empty, located in the ProgramData/Bonic/project/etc directory. Then you need to shut down Seti, wait maybe 30 seconds, then restart it. You can also try the "read config" files on the menu which might very well apply the new changes. However, it won't update any of the Bonic displayed items that might change, so I usually re-start Seti. Another way to "cope" with the "slow screen response" is to setup your preferences to suspend the gpu when the computer is "active" eg. someone is using it. You won't be able to "see" the realtime GPU loads unless you over-ride that in the menu (to make it continuous), but it will take care of making the system useable whenever you need to use it. In your case, the GPU may have enough processing power that you won't really notice a difference. The main time I have a lagy video response is when I am processing the SOG type data. The Cuda 42/50 etc don't seem to have the same issue. You can play with the amount of cpu cores you devote to each process in the app_config.xml file. For right now, I would devote 1 cpu core (or hyperthread) to each gpu process. You may find you can reduce that some and not lose production. That will allow more cpu-based data crunching. There are some other parameters you can play with to increase the cpu cycles available for cpu-based data crunching while still keeping the gpu fully busy. "experimentation required" :) The -gpu_lock is the only parameter that says "sometimes faster" or "sometimes slower" testing needed. All the others appear to "help" or at least don't hinder your production. You might want to wait, 2-3 days so you have a "baseline" to compare it to, before you apply any/all of the above. Tom A proud member of the OFA (Old Farts Association). |
Stauning Send message Joined: 10 Nov 00 Posts: 25 Credit: 67,819,196 RAC: 749 |
To summerise: If still low RAC try to delete gpu_lock And if better keep off, if worse reinsert. Then play with CPU pr task... Would the "normal" seti apps value, be a good start? Since my output still seems low. Michael. |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
>>Then play with CPU pr task... Would the "normal" seti apps value, be a good start? >>Since my output still seems low. The range for gpu to cpu runs from 1 to 1 to .5(gpu) to .134 (or there abouts) cpu. I am getting decent results with .5 gpu and .25 cpu which means no more than 1 cpu is tied up for 2 video cards running 2 tasks each. Someone else has recommend .5 gpu and .33 cpu which will tie upto 2 cpus when you have 2 video cards. If you are running the single card then you could either run .5 gpu and .25 cpu to use just one cpu for the card. Or you could run .5 gpu and 1 cpu to get 2 cpus dedicated. Unless you are getting significant load on both cpus, you might want to only use 1 cpu. All this discussion applies only to "SOG" data. If you are processing any of the Mcuda50, Mcuda42 etc data then the cpu/gpu loads will be less. If you are running 1 cpu per gpu task (probably 2 tasks / video card) and your cpu as listed in the Task manager while running "SOG" tasks is not showing upwards to 25% then it is possible you can get more production by using the -cpu_lock or -high_perf. However, if you are averaging around 30 minutes or less per task then it may not be possible to speed it up. Apparently your computer list is now private so I can't compare your results to mine. If you would take a look at my Xeon box and see if you are at least equal to my production then you aren't doing bad. You almost certainly can do better. But not order of magnitudes better (eg. 10 times) I think it would be more productive to start a new thread in the Number crunching message area if you want to get any further advice. A proud member of the OFA (Old Farts Association). |
©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.