There are few statistics, but it seems that in half of the cases, after much torment, it ends successfully.
Previously, ~80% ended with an error.
Something like this.
If the success rate goes up, that would mean that Ian&Steve C. were pretty much right when they said that not all of the RAM is available to the integrated graphics card.
There are few statistics, but it seems that in half of the cases, after much torment, it ends successfully.
Previously, ~80% ended with an error.
Something like this.
If the success rate goes up, that would mean that Ian&Steve C. were pretty much right when they said that not all of the RAM is available to the integrated graphics card.
If you take a look at the number of PC's, and the stats of those same PC's, then you would not doubt the advise of Ian&Steve C. when it comes to Boinc related issues.
I have the same issue. 5950x, 128gb RAM, Nvidia GTX 970, 12 tasks stuck at 89.978% with several others at smaller percentages complete making slow, almost non-existent progress. All tasks are for the LATeah project.
According to your tasks list, you don't have any GPU tasks at all - only CPU tasks. There are two separate searches, one for CPUs and one for GPUs (with CPU support) and you have tasks for the CPU only search (FGRP5). According to the last scheduler contact that I looked at, your host wasn't even requesting GPU tasks (FGRPB1G). Are you sure you have selected that GPU search in your preferences?
If that search is selected, the next thing to check is whether or not the BOINC client properly detects your GPU hardware and OpenCL compute libs. Look for the startup messages in the event log in BOINC Manager. Your GPU should be listed together with the version of OpenCL that is installed. I don't use Windows so I don't have examples of exactly what you should see. You need to make sure you have a working OpenCL setup. I don't know that Microsoft versions of drivers include OpenCL. From what others have mentioned previously, you might need to get that from nVidia.
Tyler Wilcock wrote:
... 12 tasks stuck at 89.978% with several others at smaller percentages complete making slow, almost non-existent progress.
Since you are only running CPU tasks, it sounds like you might be overloading your CPU. It is listed as 16 cores, 32 threads. How many tasks in total are trying to run? Hopefully not the full 32 threads?
Both types of gamma-ray pulsar searches perform the calculation in two stages. The search for candidate signals happens from 0% to ~90%. After that there is a 'followup' stage where the top ten candidates found in the main stage are re-evaluated. The followup stage can take a while (maybe 20-30 mins or even longer), depending on the strength of the CPU - particularly if the CPU is already overloaded.
If you are using all CPU threads, that might be why things go very slowly. All you need to do is use the preference setting to restrict the % of processors that BOINC can use. Start at 50% and see how much improvement that makes. Try different values until you get the most productive setting. This will be particularly important if you do start running GPU tasks as well as CPU tasks.
Thanks for the advice, and sorry for my misunderstanding. You are indeed correct that I'm only running CPU tasks, so my problem is different than OPs.
Since you are only running CPU tasks, it sounds like you might be overloading your CPU. It is listed as 16 cores, 32 threads. How many tasks in total are trying to run? Hopefully not the full 32 threads?
I was allowing BOINC to use 100% of my CPU 100% of the time, so I suppose this could be the case. It's unintuitive to me that lowering the percentage of CPU I allow BOINC to use will make tasks complete faster / more reliably, but I trust your expertise in this area over my own.
I've switched my settings such that BOINC can use 80% of my CPU 100% of the time, so let's see how that goes. So far so good.
It's unintuitive to me that lowering the percentage of CPU I allow BOINC to use will make tasks complete faster ...
CPU tasks are compute intensive and do manipulate a large amount of data in memory. You need a lot of memory when you are running all threads. Whilst you may add more memory and use higher speed memory, there will always be a limit (memory bandwidth) as to how quickly data can flow back and forth between the CPU and the memory being used. Once you exceed the available bandwidth, there tends to be a fairly drastic drop in overall efficiency.
As you increase the number of running threads, you are likely to exceed the available bandwidth well before you run out of threads on which to run extra jobs. If you keep pushing beyond that point, the overall output tends to decline significantly. To get the best overall output, you really need to experiment by starting lower than 80% and gradually working up. That's why I suggested 50% as a starting point.
The only way to be sure for your setup is to try a range of values to see what gives the best overall output.
I found a way specifically
)
I found a way specifically for Intel Graphics HD to increase the video memory limit in Windows 7.
I put it as much as 4 GB.
I'll see if it does something
Good luck, hopefully that
)
Good luck, hopefully that solves your issue!
It seems to have gotten
)
It seems to have gotten better.
There are few statistics, but it seems that in half of the cases, after much torment, it ends successfully.
Previously, ~80% ended with an error.
Something like this.
If the success rate goes up, that would mean that Ian&Steve C. were pretty much right when they said that not all of the RAM is available to the integrated graphics card.
Eliovich Alexander & Yan
)
If you take a look at the number of PC's, and the stats of those same PC's, then you would not doubt the advise of Ian&Steve C. when it comes to Boinc related issues.
I have the same issue. 5950x,
)
I have the same issue. 5950x, 128gb RAM, Nvidia GTX 970, 12 tasks stuck at 89.978% with several others at smaller percentages complete making slow, almost non-existent progress. All tasks are for the LATeah project.
Tyler Wilcock wrote:I have
)
Unfortunately, I don't think that's possible :-).
According to your tasks list, you don't have any GPU tasks at all - only CPU tasks. There are two separate searches, one for CPUs and one for GPUs (with CPU support) and you have tasks for the CPU only search (FGRP5). According to the last scheduler contact that I looked at, your host wasn't even requesting GPU tasks (FGRPB1G). Are you sure you have selected that GPU search in your preferences?
If that search is selected, the next thing to check is whether or not the BOINC client properly detects your GPU hardware and OpenCL compute libs. Look for the startup messages in the event log in BOINC Manager. Your GPU should be listed together with the version of OpenCL that is installed. I don't use Windows so I don't have examples of exactly what you should see. You need to make sure you have a working OpenCL setup. I don't know that Microsoft versions of drivers include OpenCL. From what others have mentioned previously, you might need to get that from nVidia.
Since you are only running CPU tasks, it sounds like you might be overloading your CPU. It is listed as 16 cores, 32 threads. How many tasks in total are trying to run? Hopefully not the full 32 threads?
Both types of gamma-ray pulsar searches perform the calculation in two stages. The search for candidate signals happens from 0% to ~90%. After that there is a 'followup' stage where the top ten candidates found in the main stage are re-evaluated. The followup stage can take a while (maybe 20-30 mins or even longer), depending on the strength of the CPU - particularly if the CPU is already overloaded.
If you are using all CPU threads, that might be why things go very slowly. All you need to do is use the preference setting to restrict the % of processors that BOINC can use. Start at 50% and see how much improvement that makes. Try different values until you get the most productive setting. This will be particularly important if you do start running GPU tasks as well as CPU tasks.
Cheers,
Gary.
Thanks for the advice, and
)
Thanks for the advice, and sorry for my misunderstanding. You are indeed correct that I'm only running CPU tasks, so my problem is different than OPs.
I was allowing BOINC to use 100% of my CPU 100% of the time, so I suppose this could be the case. It's unintuitive to me that lowering the percentage of CPU I allow BOINC to use will make tasks complete faster / more reliably, but I trust your expertise in this area over my own.
I've switched my settings such that BOINC can use 80% of my CPU 100% of the time, so let's see how that goes. So far so good.
Tyler Wilcock wrote:It's
)
CPU tasks are compute intensive and do manipulate a large amount of data in memory. You need a lot of memory when you are running all threads. Whilst you may add more memory and use higher speed memory, there will always be a limit (memory bandwidth) as to how quickly data can flow back and forth between the CPU and the memory being used. Once you exceed the available bandwidth, there tends to be a fairly drastic drop in overall efficiency.
As you increase the number of running threads, you are likely to exceed the available bandwidth well before you run out of threads on which to run extra jobs. If you keep pushing beyond that point, the overall output tends to decline significantly. To get the best overall output, you really need to experiment by starting lower than 80% and gradually working up. That's why I suggested 50% as a starting point.
The only way to be sure for your setup is to try a range of values to see what gives the best overall output.
Cheers,
Gary.