Quick Question: O3ASHF1 - does it use more than 1 CPU (core, multithreaded, or hyperthreaded) ?
I'm trying to understand utilization, and am sometimes seeing utilitization implying it may be peaking at 3 cpus in use. I've optimized the GPU side, but am seeing run time vs cpu time that is not what I expect, and think I may have resource contention.
The app is technically multithreaded, but the second thread is just there to handle the communication with the BOINC client, it shouldn't use any noticeable CPU time. All scientific processing is done in a single (CPU) thread.
Quick Question: O3ASHF1 - does it use more than 1 CPU (core, multithreaded, or hyperthreaded) ?
I'm trying to understand utilization, and am sometimes seeing utilitization implying it may be peaking at 3 cpus in use. I've optimized the GPU side, but am seeing run time vs cpu time that is not what I expect, and think I may have resource contention.
Since you have a fast cpu (13900K), the cpu calculation portion is taking longer than what I'm expecting. See my old post here for linux system with AMD card but it should also be applicable for window system with Nvidia card. Are you running any other cpu projects also? I found that when I run cpu boinc projects with O3AS at almost 100% cpu utilization, I notice the cpu portion run time started to increase significantly. Suggest trying with less cpu total utilization and see how it goes first. Note if you set your cpu at default stock and only run O3AS in your system the cpu core clock may turbo boost to higher frequency when few cores are running, so may see a slight bump in speed vs running more cores.
One of your tasks such as this that you ran, the cpu calculation time took almost 13 mins (from 04:32:03 to 04:45:14). That's pretty long for 13900K.
2024-07-08 04:32:03.3487 (13400) [normal]: Finished main analysis.
2024-07-08 04:32:03.3487 (13400) [normal]: Recalculating statistics for the final toplist(s)...
2024-07-08 04:45:14.3890 (13400) [normal]: Finished recalculating toplist statistics.
2024-07-08 04:45:14.3890 (13400) [normal]: Finished in 1385.58 s with peak RAM usage: -1.0 MB on CPU '13th Gen Intel(R) Core(TM) i9-13900K'
Note: you can try to stagger each task when running multiple tasks per gpu but at some point in time, a few tasks will lose their stagger. Not sure in windows but on my linux system, I run 6 tasks per gpu and most of the time I lose stagger on 2 or 3 tasks only.
Since you have a fast cpu (13900K), the cpu calculation portion is taking longer than what I'm expecting. See my old post here for linux system with AMD card but it should also be applicable for window system with Nvidia card.
Yes, I agree, CPU calc is taking quite a long time. Thanks for the suggestions. I'm going to play with the load and the tasks processing to free up CPU and see what is actually going on. Am wondering if cache is getting used up by one of the non-GPU tasks running. I'll dial back CPU avaialbility and try to eliminate CPU-only tasks and see what happens.
pututu wrote:
Note: you can try to stagger each task when running multiple tasks per gpu but at some point in time, a few tasks will lose their stagger. Not sure in windows but on my linux system, I run 6 tasks per gpu and most of the time I lose stagger on 2 or 3 tasks only.
Yeah, I found the same thing, after a while things sync'd up and so lost the advantage.
For O3ASHF1d there is work for 13 days only. That's pretty soon. Does it go on in any way?
Kind regards and happy crunching
Martin
Like I said in your other post (Einstein@home >> Forums >> about similar tasks, BRP4G and FGRP5, this is really no different.
As you can see, this shows a time period of one month where O3AS is dropping to virtually zero tasks, rarely actually zero, and then rebounds up to another fulfillment.
For O3ASHF1d there is work for 13 days only. That's pretty soon. Does it go on in any way?
Kind regards and happy crunching
Martin
Short term: as explained by GWGeorge007.
Long term: This specific O3AS batch will continue until the frequency reaches 2000Hz. From what I can find the latest task issued is currently 1570.00Hz, so there is still a long way to go for this batch to reach its goal. We started at 1500.Hz back in late May, so in about 3 months or so, we have progressed by 70Hz.
Long term: This specific O3AS batch will continue until the frequency reaches 2000Hz. From what I can find the latest task issued is currently 1570.00Hz, so there is still a long way to go for this batch to reach its goal. We started at 1500.Hz back in late May, so in about 3 months or so, we have progressed by 70Hz.
We are not even half way there yet...
If only one of the big GPU's cluster from Elon could help... :)
tasks are flowing again
)
tasks are flowing again
_________________________________________________________________________
Quick Question: O3ASHF1 -
)
Quick Question: O3ASHF1 - does it use more than 1 CPU (core, multithreaded, or hyperthreaded) ?
I'm trying to understand utilization, and am sometimes seeing utilitization implying it may be peaking at 3 cpus in use. I've optimized the GPU side, but am seeing run time vs cpu time that is not what I expect, and think I may have resource contention.
The app is technically
)
The app is technically multithreaded, but the second thread is just there to handle the communication with the BOINC client, it shouldn't use any noticeable CPU time. All scientific processing is done in a single (CPU) thread.
BM
Colin Haig wrote:Quick
)
Since you have a fast cpu (13900K), the cpu calculation portion is taking longer than what I'm expecting. See my old post here for linux system with AMD card but it should also be applicable for window system with Nvidia card. Are you running any other cpu projects also? I found that when I run cpu boinc projects with O3AS at almost 100% cpu utilization, I notice the cpu portion run time started to increase significantly. Suggest trying with less cpu total utilization and see how it goes first. Note if you set your cpu at default stock and only run O3AS in your system the cpu core clock may turbo boost to higher frequency when few cores are running, so may see a slight bump in speed vs running more cores.
One of your tasks such as this that you ran, the cpu calculation time took almost 13 mins (from 04:32:03 to 04:45:14). That's pretty long for 13900K.
Note: you can try to stagger each task when running multiple tasks per gpu but at some point in time, a few tasks will lose their stagger. Not sure in windows but on my linux system, I run 6 tasks per gpu and most of the time I lose stagger on 2 or 3 tasks only.
pututu wrote: Since you have
)
Yes, I agree, CPU calc is taking quite a long time. Thanks for the suggestions. I'm going to play with the load and the tasks processing to free up CPU and see what is actually going on. Am wondering if cache is getting used up by one of the non-GPU tasks running. I'll dial back CPU avaialbility and try to eliminate CPU-only tasks and see what happens.
Yeah, I found the same thing, after a while things sync'd up and so lost the advantage.
Hallo! For O3ASHF1d there
)
Hallo!
For O3ASHF1d there is work for 13 days only. That's pretty soon. Does it go on in any way?
Kind regards and happy crunching
Martin
astro-marwil
)
Like I said in your other post (Einstein@home >> Forums >> about similar tasks, BRP4G and FGRP5, this is really no different.
As you can see, this shows a time period of one month where O3AS is dropping to virtually zero tasks, rarely actually zero, and then rebounds up to another fulfillment.
Again, I hope this helps.
Proud member of the Old Farts Association
astro-marwil
)
EDIT: Duplicate post deleted.
Proud member of the Old Farts Association
astro-marwil
)
Short term: as explained by GWGeorge007.
Long term: This specific O3AS batch will continue until the frequency reaches 2000Hz. From what I can find the latest task issued is currently 1570.00Hz, so there is still a long way to go for this batch to reach its goal. We started at 1500.Hz back in late May, so in about 3 months or so, we have progressed by 70Hz.
Quote:Quote: Long term: This
)
We are not even half way there yet...
If only one of the big GPU's cluster from Elon could help... :)