I have not had much luck running multiple Einstein tasks on a single gpu. While some improvement in elapsed time is possible on faster cards the heat becomes a factor as utilization goes up.
The system with that 1050 has an intel GPU that can also run Einstein. Have you tried running the Intel GPU app?
You could try out 2x at first... by setting 0.5 for "GPU utilization factor" (or configuring app_config.xml in a similar manner). Your GTX 1050 Ti has 4GB memory so it would be able to run two GW tasks concurrently.
If you'd like to run FGRPB1G tasks then that card would be okay for 3x (GPU utilization factor of 0.33). That might give a very small additional boost for credit output if compared with 2x.
Your computer has plenty of cores available so the GPU will probably get a sufficient amount of CPU support.
I think 2x could be good in general, because it would cause errors less likely than 3x and the additional benefit from running 3x tends to be quite small for these middle category Nvidia cards. But you could test both and see how they run and then make the decision. Just monitor the GPU temp. For example MSI Afterburner or GPU-Z software is good for displaying some graphs at how the card is loaded while running.
[edit] the above is the app_config.xml file that Richie mentioned. It goes into Boinc/Projects/Einstein..folder
Change the 0.75 to 1.0 as you have plenty of CPU cores.
With the 0.5 you will run two at a time. Run for a day to get some statistics then change the 0.5 to 0.333 for three concurrent tasks then try 0.25 to get 4 at a time.
Avg: 2366.5 2262.2 1000.0
STD: 448.7 403.5 0.0
2.37 seconds per credit from above info one device
0.4226 Credits per second for one device
Times shown above were divided by number of concurrent tasks (1)
I did not know. I will try right now... Thank you!
You've actually been running FGRPB1G tasks on your Intel GPU for some days now, apparently without realizing this :-). For example, this task was returned on 6th Feb with a crunch time of nearly 3 hours. By looking closely at the task ID link, you can see the app has a suffix of "(FGRPopencl-intel_gpu)" which tells you the device used to crunch the task. If you go wildly setting multiple concurrent tasks, those tasks on the Intel GPU may very well cause problems.
I'm certainly not the best person to advise you how best to use your hardware. I run Linux only, I don't use Intel or nvidia GPUs (only AMD) and I don't use mobile devices, so I have no relevant experience. Also you have lots of CPU (12 threads - which is useful) but the speed is listed as 2.2GHz which suggests the CPU is crippled (frequency-wise) to reduce heat production. I guess your device is a mobile device, perhaps with a fairly limited cooling system and therefore not really suited to heavy duty crunching. If you don't want to shorten its life, you should be very careful with how heavily you load it up with crunching. You will create a lot of heat by using both your GPUs.
Also, you are running the gravity wave (GW) GPU app, the Gamma-ray pulsar (GRP) GPU app and the Arecibo radio pulsar tasks, most of which you have aborted. If you don't want the Arecibo stuff, you should disable that search in your project preferences.
The GW stuff is running on the 1050Ti and it probably would benefit from running a second concurrent task. If you run both GW and GRP work simultaneously, you will create more heat and the two search types may well interfere with each other by causing estimated crunch times to oscillate, sometimes quite wildly. Until you get more experience, you would be well advised to select just one of these search types. You can also select or deselect the different GPU types. You should takes things very slowly and experiment with each search separately until you find what works best for you.
... the above is the app_config.xml file that Richie mentioned. It goes into Boinc/Projects/Einstein..folder
Hey guys, please realize that this is the "Getting Started" forum so unless you are prepared to supply a *lot* of ongoing support, you are probably not well advised to be suggesting stuff that even experienced users often have trouble with.
The first thing to sort out is what search (or searches) are preferred and how best to run single tasks for that search. The choice of search will certainly impact on how many concurrent tasks might be recommended when it's time to think about concurrency.
Well, I wanted to leave the other person the joy of finding a solution, piece by piece. Even experimenting. My initial quess is that he may have gathered some information on this area and if not quite enough to have every detail yet then additional questions and answers will lead forward and are always welcome.
You can either set it in an app_info or app_config file.
Or the easiest method is by the Project Preferences page on the website. On the bottom of the page is a "GPU Utilization factor" for each type of gpu app sub-project.
The default is 1.0 and if you change it to 0.5 you will run two concurrent tasks on each gpu. 0.33 for three concurrent tasks. Same idea as Seti.
Be aware that some apps respond better for multiple task concurrency and others not so well. You will have to test for yourself. AMD cards respond well and Nvidia cards not so much.
I have not had much luck
)
I have not had much luck running multiple Einstein tasks on a single gpu. While some improvement in elapsed time is possible on faster cards the heat becomes a factor as utilization goes up.
The system with that 1050 has an intel GPU that can also run Einstein. Have you tried running the Intel GPU app?
I did not know. I will try
)
I did not know. I will try right now... Thank you!
You could try out 2x at
)
You could try out 2x at first... by setting 0.5 for "GPU utilization factor" (or configuring app_config.xml in a similar manner). Your GTX 1050 Ti has 4GB memory so it would be able to run two GW tasks concurrently.
If you'd like to run FGRPB1G tasks then that card would be okay for 3x (GPU utilization factor of 0.33). That might give a very small additional boost for credit output if compared with 2x.
Your computer has plenty of cores available so the GPU will probably get a sufficient amount of CPU support.
I think 2x could be good in general, because it would cause errors less likely than 3x and the additional benefit from running 3x tends to be quite small for these middle category Nvidia cards. But you could test both and see how they run and then make the decision. Just monitor the GPU temp. For example MSI Afterburner or GPU-Z software is good for displaying some graphs at how the card is loaded while running.
https://einsteinathome.org/co
)
https://einsteinathome.org/content/set-cpu-usage-gpu#comment-175569
[edit] the above is the app_config.xml file that Richie mentioned. It goes into Boinc/Projects/Einstein..folder
Change the 0.75 to 1.0 as you have plenty of CPU cores.
With the 0.5 you will run two at a time. Run for a day to get some statistics then change the 0.5 to 0.333 for three concurrent tasks then try 0.25 to get 4 at a time.
Use this tool to get statistics https://stateson.net.HostProjectStats
Your current statistics show the following
The above came from this url
https://stateson.net/HostProjectStats/default.aspx?url=https://einsteinathome.org/host/12804960/tasks/4/54;nCon=1;nDev=1;Wu=20;lw=0;iw=0;
click on it and then select "Calculate"
when running more than one concurrent task change nCon from 1 to 2 or 3 or whatever fits in the NVidia card.
NOGOOD wrote:I did not know.
)
You've actually been running FGRPB1G tasks on your Intel GPU for some days now, apparently without realizing this :-). For example, this task was returned on 6th Feb with a crunch time of nearly 3 hours. By looking closely at the task ID link, you can see the app has a suffix of "(FGRPopencl-intel_gpu)" which tells you the device used to crunch the task. If you go wildly setting multiple concurrent tasks, those tasks on the Intel GPU may very well cause problems.
I'm certainly not the best person to advise you how best to use your hardware. I run Linux only, I don't use Intel or nvidia GPUs (only AMD) and I don't use mobile devices, so I have no relevant experience. Also you have lots of CPU (12 threads - which is useful) but the speed is listed as 2.2GHz which suggests the CPU is crippled (frequency-wise) to reduce heat production. I guess your device is a mobile device, perhaps with a fairly limited cooling system and therefore not really suited to heavy duty crunching. If you don't want to shorten its life, you should be very careful with how heavily you load it up with crunching. You will create a lot of heat by using both your GPUs.
Also, you are running the gravity wave (GW) GPU app, the Gamma-ray pulsar (GRP) GPU app and the Arecibo radio pulsar tasks, most of which you have aborted. If you don't want the Arecibo stuff, you should disable that search in your project preferences.
The GW stuff is running on the 1050Ti and it probably would benefit from running a second concurrent task. If you run both GW and GRP work simultaneously, you will create more heat and the two search types may well interfere with each other by causing estimated crunch times to oscillate, sometimes quite wildly. Until you get more experience, you would be well advised to select just one of these search types. You can also select or deselect the different GPU types. You should takes things very slowly and experiment with each search separately until you find what works best for you.
Cheers,
Gary.
JStateson wrote:... the above
)
Hey guys, please realize that this is the "Getting Started" forum so unless you are prepared to supply a *lot* of ongoing support, you are probably not well advised to be suggesting stuff that even experienced users often have trouble with.
The first thing to sort out is what search (or searches) are preferred and how best to run single tasks for that search. The choice of search will certainly impact on how many concurrent tasks might be recommended when it's time to think about concurrency.
Cheers,
Gary.
Well, I wanted to leave the
)
Well, I wanted to leave the other person the joy of finding a solution, piece by piece. Even experimenting. My initial quess is that he may have gathered some information on this area and if not quite enough to have every detail yet then additional questions and answers will lead forward and are always welcome.
Been working SETI for years
)
Been working SETI for years which has gone inactive so applying resources to this project. How do I set the GPU factor?
You can either set it in an
)
You can either set it in an app_info or app_config file.
Or the easiest method is by the Project Preferences page on the website. On the bottom of the page is a "GPU Utilization factor" for each type of gpu app sub-project.
The default is 1.0 and if you change it to 0.5 you will run two concurrent tasks on each gpu. 0.33 for three concurrent tasks. Same idea as Seti.
Be aware that some apps respond better for multiple task concurrency and others not so well. You will have to test for yourself. AMD cards respond well and Nvidia cards not so much.