Message boards :
Number crunching :
Setting up Linux to crunch CUDA90 and above for Windows users
Message board moderation
Previous · 1 . . . 106 · 107 · 108 · 109 · 110 · 111 · 112 . . . 162 · Next
Author | Message |
---|---|
Siran d'Vel'nahr Send message Joined: 23 May 99 Posts: 7379 Credit: 44,181,323 RAC: 238 |
B-B-R-O-Y-G-B-V-G-W lol :) I only know the R version, never heard a G version. Hi Keith, Back in high school, in the 60s, I took an electronics class. I built a tube type AM radio that actually worked. My dad told me about the resistor color code and said it like this: "Bad Boys R___ Our Young Girls Behind Victory Garden Walls" He never told me a G rated version. lol :) Have a great day! :) Siran CAPT Siran d'Vel'nahr - L L & P _\\// Winders 11 OS? "What a piece of junk!" - L. Skywalker "Logic is the cement of our civilization with which we ascend from chaos using reason as our guide." - T'Plana-hath |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
With the 2080 Ti, you should install the 0.98b1 CUDA101 application as that would work better than the stock CUDA90 application that is provided in the app_info. The application is already in the project directory. All you need to do is a Find and Replace with the Text Editor in the app_info for the CUDA101 appname for the CUDA90 appname. You are running the CUDA101 application now. But still not using all of the capability of the cards. You need to add the -nobs parameter to the <cmdline></cmdline> statement in the app_info for the CUDA app version section. That will allow the CUDA thread to use an entire cpu thread to support the task. It will shave off another 5 seconds or so of compute time. The compute_time and the run_time will then match up in the stderr.txt output for each task. It will drive the utilization up on the cards and they will run a lot closer to the 250W TDP of the card. I hope you have enough power supply and cooling to handle that. You can also restrict the power usage of the cards via nvidia-smi and take the power usage down to 225 or 215W and it won't affect the crunch time all that much. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
If you are just trying to download the All-in-One installer, you don't need to sign into Crunchers Anonymous for that. Here is the direct download link. ++1! A proud member of the OFA (Old Farts Association). |
elec999 Send message Joined: 24 Nov 02 Posts: 375 Credit: 416,969,548 RAC: 141 |
I started moving all my crunches to Linux. Lots to learn. But thank you guys for the help and support. Much appreciated |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
You haven't added the -nobs parameter yet to your cmdline statement. That will speed the crunch times up quite a bit. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Juhani Karjanlahti Send message Joined: 23 Jan 03 Posts: 15 Credit: 83,675,733 RAC: 149 |
You haven't added the -nobs parameter yet to your cmdline statement. That will speed the crunch times up quite a bit. There is still something else slowing his machine down. I checked his runtimes https://setiathome.berkeley.edu/workunit.php?wuid=3458788392 <- this WU is validated against my linux box and it suggests my 2060 is 25 seconds faster than his 2080 ti. I know my test version cuda 10.0 app is 5-7% faster than 10.1 but that 30% speed difference should be other way around. |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
He could be overcommitted on cpu tasks and not giving the gpu thread enough time slices. The -nobs parameter will fix that. The other possibility is he doesn't have enough power supply to support 4 X 250 watts of RTX 2080Ti's. He could be overheating them too if he doesn't have enough cooling. The gpu clocks could be way down if power or thermal limited. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
elec999 Send message Joined: 24 Nov 02 Posts: 375 Credit: 416,969,548 RAC: 141 |
You haven't added the -nobs parameter yet to your cmdline statement. That will speed the crunch times up quite a bit. Can you give me some help with this "-nobs" Thank you |
Ian&Steve C. Send message Joined: 28 Sep 99 Posts: 4267 Credit: 1,282,604,591 RAC: 6,640 |
honestly i think with a slowdown like that, it's more than just the lack of -nobs. -nobs usually only adds a small boost, not double. from his previous screenshot, it looks like his GPUs are only seeing 50-60% utilization. that could be because the CPU is too busy to feed them, or it could be a USB cable or riser issue (if he's using them). elec, if you are not using any kind of risers, what CPU percentage do you have set in your preferences? if you have it set to 100%, try restricting it to 80 or 90%. you need to free up CPU resources to feed the GPUs. from the BOINC Manager Advanced View -> Options -> Computing preferences -> Computing tab under Usage limits: Use at most ____ % of the CPUs -> change this to 80% Seti@Home classic workunits: 29,492 CPU time: 134,419 hours |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
You haven't added the -nobs parameter yet to your cmdline statement. That will speed the crunch times up quite a bit. First do as Ian suggests and reduce the cpu usage down to 80% so there is enough cpu support for the gpus. The -nobs parameter gets added to the <cmdline>-nobs</cmdline> statement position in either the app_config.xml or app_info.xml. That forces the app to use a full cpu thread to support each gpu thread. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
elec999 Send message Joined: 24 Nov 02 Posts: 375 Credit: 416,969,548 RAC: 141 |
The cards are directly into the motherboard pciexpress slots. The PSU is 1500watt evga. |
Ian&Steve C. Send message Joined: 28 Sep 99 Posts: 4267 Credit: 1,282,604,591 RAC: 6,640 |
ok, so ignore what i said about risers, but still reduce your CPU use like I outlined above. did it help? Seti@Home classic workunits: 29,492 CPU time: 134,419 hours |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
ok, so ignore what i said about risers, but still reduce your CPU use like I outlined above. Looks like it. Either/both the reduction in cpu usage or the addition of the -nobs parameter got the task time down to what a 2080 Ti should be running on the 0.98b1 CUDA101 application. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Ian&Steve C. Send message Joined: 28 Sep 99 Posts: 4267 Credit: 1,282,604,591 RAC: 6,640 |
yup. looks like he's back on track. Seti@Home classic workunits: 29,492 CPU time: 134,419 hours |
elec999 Send message Joined: 24 Nov 02 Posts: 375 Credit: 416,969,548 RAC: 141 |
Thanks guys Would this special app work with dual 690s? |
Wiggo Send message Joined: 24 Jan 00 Posts: 34887 Credit: 261,360,520 RAC: 489 |
Thanks guysNo, you at least need a Maxwell based GPU. Cheers. |
elec999 Send message Joined: 24 Nov 02 Posts: 375 Credit: 416,969,548 RAC: 141 |
Thanks guysNo, you at least need a Maxwell based GPU. Would Linux have any benifits to these gpus? |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
Thanks guysNo, you at least need a Maxwell based GPU. Possibly. The 690's could run the SoG application but the Linux CUDA60 zi3v application may still be faster. TBar made that available at Crunchers too. http://www.arkayn.us/lunatics/CUDA60_zi3v-Special.7z After further thought and investigation, I don't think the 690 is capable of running the special app after all because I believe the cutoff is CC 3.5 and the 690 only has CC 3.0. Hope someone confirms my thinking. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
Thanks guysNo, you at least need a Maxwell based GPU. . . I thought it was cc=3.2 but either way these cards don't have it. Stephen :( |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
Well I just downloaded the archive and you are correct Stephen. From the readme: "This App will only work on GPUs that are Compute Capability 3.2 or higher." Which is moot because the Wikipedia entry shows there never was any card actually built with the GK20A die. The next higher capability die to be used was CC 3.5 capable. So my original post was actually accurate. Seti@Home classic workunits:20,676 CPU time:74,226 hours 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.