Message boards :
Number crunching :
GTX780 Ti system
Message board moderation
Previous · 1 · 2 · 3 · Next
Author | Message |
---|---|
Ghia Send message Joined: 7 Feb 17 Posts: 238 Credit: 28,911,438 RAC: 50 |
I would suggest that you not do that at this time. I've burned out 2 780Tis a while back by increasing the number of instances of work on them. They held out for a while but no one likes that smell of burnt electronics when they come home. Thank you ! I was just thinking a GTX 780Ti Dual Classified should be able to do more... A question : Is the file mb_cmdline-8.22_intel_opencl_nvidia_SoG.txt (in projects dir.) generated by the application ? Should it be empty ? Humans may rule the world...but bacteria run it... |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
A question : Is the file mb_cmdline-8.22_intel_opencl_nvidia_SoG.txt (in projects dir.) generated by the application ? Should it be empty ? It is part of the OpenCL_SoG package and yes it should be empty by default. For optimisation users can add app args to fine tune the application. I was just thinking a GTX 780Ti Dual Classified should be able to do more... It certainy is. If your temps are O.K. i would suggest to consider installing Lunatics optimised apps and running 2 instances on the GPU. With each crime and every kindness we birth our future. |
Brent Norman Send message Joined: 1 Dec 99 Posts: 2786 Credit: 685,657,289 RAC: 835 |
It is not auto generated, but sent to you by the server to place customized command line parameters that you want to use. You may or not also have an app_config.xml file as well with you would use to run more than 1 task at a time. For my 750Ti I use: -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 -hpfor the command line, 1 task at a time. EDIT: I don't feel that running 2 tasks on my card is an advantage since it ties up 2 cores for very little GPU increase. |
Ghia Send message Joined: 7 Feb 17 Posts: 238 Credit: 28,911,438 RAC: 50 |
It is part of the OpenCL_SoG package and yes it should be empty by default. Understood... Right after the time I was able to take the SoG application off of my Firewall blacklist, a stderr.txt file was created. It looks like this : 09:24:17 (1964): Can't open init data file - running in standalone mode 09:24:17 (1964): Can't open init data file - running in standalone mode Not using mb_cmdline.txt-file, using commandline options. Priority of worker thread raised successfully Priority of process adjusted successfully, below normal priority class used 09:24:17 (1964): Can't open init data file - running in standalone mode WARNING: init_data.xml missing OpenCL platform detected: Advanced Micro Devices, Inc. OpenCL platform detected: NVIDIA Corporation WARNING: BOINC supplied wrong platform! BOINC assigns device 0 WARNING: BOINC failed to provide OpenCL device, using own enumeration abilities SETI@home error -5 Can't open file (work_unit.sah) in read_wu_state() errno=2 File: ..\worker.cpp Line: 136 Would this have something to do with the SoG application not running ? I don't have the knowledge to understand what this actually means... Is a project reset in order ? Humans may rule the world...but bacteria run it... |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
SETI@home error -5 Can't open file It seems its still blocked somewhere. With each crime and every kindness we birth our future. |
Brent Norman Send message Joined: 1 Dec 99 Posts: 2786 Credit: 685,657,289 RAC: 835 |
For SoG to run, you need, an EXE file with SoG in it's name, and a MultiBeam CL file with the same r#### as the EXE file. You might want to restart BOINIC and watch for errors in ~ the first 30 lines and post them here so that we can see it. BOINC 'should' complain if needed files are missing and try to get them from the server if/when needed (unless you manually installed the SETI apps). It sure seems like you are stuck on CUDA42 at the moment. |
Brent Norman Send message Joined: 1 Dec 99 Posts: 2786 Credit: 685,657,289 RAC: 835 |
Wait a minute, Did you just whitelist the EXE file in your anti virus? The EXE will create WISDOM and BIN files that it creates specific to your card before it runs. If those are getting blocked by your AV it will not run. Whitelist the entire directory I.E. *.* |
Ghia Send message Joined: 7 Feb 17 Posts: 238 Credit: 28,911,438 RAC: 50 |
Wait a minute, I whitelisted the whole project directory on Feb.11. The downloaded SoG units errored out long before that, of course. The quoted stderr.txt appeared AFTER I whitelisted the project diretory. HTH... Humans may rule the world...but bacteria run it... |
Ghia Send message Joined: 7 Feb 17 Posts: 238 Credit: 28,911,438 RAC: 50 |
You might want to restart BOINIC and watch for errors in ~ the first 30 lines and post them here so that we can see it. Will do...any flags to enable apart from the default ones ? Humans may rule the world...but bacteria run it... |
Ghia Send message Joined: 7 Feb 17 Posts: 238 Credit: 28,911,438 RAC: 50 |
Just a small update : I did a project reset when the servers came online after yesterday's outage. No WUs for the SoG application so far, so the files for it are absent. For some reason, crunching times have gone up significantly...dunno if this is coincidence or not. GPU load is down and CPU load is up, can't explain why. Humans may rule the world...but bacteria run it... |
Ghia Send message Joined: 7 Feb 17 Posts: 238 Credit: 28,911,438 RAC: 50 |
Chose to post here instead of making a new thread, hope that's ok. In app_config, I initially set my GPU to run 2 tasks on cuda42, by setting : <app_config> <app_version> <app_name>setiathome_v8</app_name> <plan_class>cuda42</plan_class> <avg_ncpus>0.2</avg_ncpus> <ngpus>0.5</ngpus> </app_version> </app_config> That worked just fine. But then the SoG-app did not run any tasks until I changed it back to 1.0 GPUs and chewed through the cuda42's. So I added a section in the app_config for the SoG-app, so it looks like this : <app_config> <app> <name>setiathome_v8</name> <gpu_versions> <gpu_usage>1.0</gpu_usage> </gpu_versions> </app> <app_version> <app_name>setiathome_v8</app_name> <plan_class>opencl_nvidia_SoG</plan_class> <ngpus>1.0</ngpus> </app_version> <app> <name>setiathome_v8</name> <gpu_versions> <gpu_usage>0.5</gpu_usage> </gpu_versions> </app> <app_version> <app_name>setiathome_v8</app_name> <plan_class>cuda42</plan_class> <ngpus>0.5</ngpus> </app_version> </app_config> Will Boinc now be able to run both apps in normal order (mostly FIFO, I guess), without me having to interfere...or have I made a mess of it ? Apps are stock, and I'm running 75% of CPU and 75% of CPU time. Humans may rule the world...but bacteria run it... |
Brent Norman Send message Joined: 1 Dec 99 Posts: 2786 Credit: 685,657,289 RAC: 835 |
This might work better for you <app_config> <app> <name>setiathome_v8</name> <gpu_versions> <gpu_usage>0.5</gpu_usage> <cpu_usage>1.0</cpu_usage> </gpu_versions> </app> <app> <name>astropulse_v7</name> <gpu_versions> <gpu_usage>1</gpu_usage> <cpu_usage>1.0</cpu_usage> </gpu_versions> </app> </app_config> Nice and simple and will run them both FIFO |
Ghia Send message Joined: 7 Feb 17 Posts: 238 Credit: 28,911,438 RAC: 50 |
This might work better for you Thank you :) Could you explain a little on how this would work ? I'm mostly blank when it comes to AP....only ever had 1 of those. Does it mean the GPU will run 2 cuda42-tasks OR 1 SoG-task ? Humans may rule the world...but bacteria run it... |
Brent Norman Send message Joined: 1 Dec 99 Posts: 2786 Credit: 685,657,289 RAC: 835 |
It doesn't specify cuda/SoG so it will run either 0.5 GPU = Use 1/2 the GPU per task, so 2 tasks per card. 1.0 CPU = reserve (shutdown) 1 core per GPU task running, so 2 cores which you will need if running SoG BTW, did you ever get SOG tasks to start downloading again, or still stuck on cuda 42? |
Ghia Send message Joined: 7 Feb 17 Posts: 238 Credit: 28,911,438 RAC: 50 |
It doesn't specify cuda/SoG so it will run either I understand. What I wanted, was to run either 2 cuda42-tasks OR 1 SoG-task. That way, I can keep the 6 CPU tasks running concurrently. At least, that was the idea behind the config I posted earlier. Now Boinc hasn't touched any of the available SoG tasks yet...it chugs away at 2 cuda tasks at a time besides the 6 CPU tasks. Maybe this is the wrong way to use my resources ? BTW, did you ever get SOG tasks to start downloading again, or still stuck on cuda 42? Yes, got some SoG tasks very early this morning (and again some this afternoon), and they ran fine as described above (as a byproduct of changing the setting for cuda42 to 1GPU). Humans may rule the world...but bacteria run it... |
Ghia Send message Joined: 7 Feb 17 Posts: 238 Credit: 28,911,438 RAC: 50 |
The suggested configuration cut back to 4 CPU tasks, as expected. In addition, it will run 2 cuda tasks, OR 1 cuda task and 1 SoG task. The GPU load gets a bit high when using 1.5 GPU (?) as in 1 cuda and 1 SoG task concurrently. To me, that looks like a setback, when compared to 6 CPU tasks and 2 cuda tasks (OR 6 CPU tasks and 1 SoG task). Running times seem pretty much the same as before, but of course it is hard to say with any certainty without lots more data. I feel I should be able to have my cake and it eat it too, I guess (i.e. 6 CPU tasks, plus 2 cuda tasks OR 1 SoG task). Humans may rule the world...but bacteria run it... |
Michel Makhlouta Send message Joined: 21 Dec 03 Posts: 169 Credit: 41,799,743 RAC: 0 |
I was running 3 cuda 42/50 on my 780's back in the day when I still had them. I used gpu-z to check the gpu load and they were nowhere near 100% on 1 WU. |
rob smith Send message Joined: 7 Mar 03 Posts: 22205 Credit: 416,307,556 RAC: 380 |
Running with 1.5 GPU per task will not result in more than 1 GPU being used for a task - the applications is not built to handle the required inter-GPU data and process transfers. SoG presents a very "lumpy" load on both the CPU and GPU, thus it is best to be a little conservative when deciding how many tasks to run at a time. Your assumption of 2 per GPU may be OK, given your CPU with 1 core (thread) free for each GPU tasks actually being run. The current applications are much more demanding on the GPU than the v7 applications were, thus the days of 3 or more tasks per GPU have passed. Even my GTX1080 struggled with v8 tasks running, but it was very comfortable running 3 tasks with the v7 applications. Bob Smith Member of Seti PIPPS (Pluto is a Planet Protest Society) Somewhere in the (un)known Universe? |
Ghia Send message Joined: 7 Feb 17 Posts: 238 Credit: 28,911,438 RAC: 50 |
I was running 3 cuda 42/50 on my 780's back in the day when I still had them. I used gpu-z to check the gpu load and they were nowhere near 100% on 1 WU. Yes, using GPU-Z to keep an eye on the numbers. Running 2 cuda tasks doesn't stress GPU load, temperature or memory controller. The SoG app is much more taxing. My card hasn't worked this hard since I bought it over 3 years ago.. ;-) Humans may rule the world...but bacteria run it... |
Ghia Send message Joined: 7 Feb 17 Posts: 238 Credit: 28,911,438 RAC: 50 |
Running with 1.5 GPU per task will not result in more than 1 GPU being used for a task - the applications is not built to handle the required inter-GPU data and process transfers. That was mye guess too...the 1.5 GPU being a "false" number. Observations here indicate that both CPU and GPU are okay with 2 cuda tasks. I haven't tried 2 SoG tasks, and probably won't anytime soon. The current applications are much more demanding on the GPU than the v7 applications were, thus the days of 3 or more tasks per GPU have passed. Even my GTX1080 struggled with v8 tasks running, but it was very comfortable running 3 tasks with the v7 applications. That's interesting...does it mean that the v8 applications have higher demands on accuracy ? This Seti interest has seen the upgrade monster raise its ugly head here....maybe it is time to build a new rig ;-) Humans may rule the world...but bacteria run it... |
©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.