Fewer GPU units please

Questions and Answers : GPU applications : Fewer GPU units please
Message board moderation

To post messages, you must log in.

AuthorMessage
Beamer-1

Send message
Joined: 19 May 99
Posts: 9
Credit: 25,250,186
RAC: 17
Message 1799295 - Posted: 29 Jun 2016, 4:23:19 UTC

I'm running an Intel i7 w/8cores, but all my WUs are CUDA/GPU and so my CPU is doing next to nothing while my GPU won't do any WUs while I'm using the computer..

Is there a setting or preference I can alter to correct this imbalance?
ID: 1799295 · Report as offensive
Profile Stubbles
Volunteer tester
Avatar

Send message
Joined: 29 Nov 99
Posts: 358
Credit: 5,909,255
RAC: 0
Canada
Message 1799417 - Posted: 29 Jun 2016, 18:26:32 UTC - in response to Message 1799295.  

Hey Beamer,

I haven't come across this specific situation before so I'll need more info.
Others with experience with such scenarios might have a quicker answer. If so, please follow their instructions.

Is this situation you are describing something that only started occurring recently? Have you made any changes to your settings recently?
From what you are describing, there seems to be 2 issues:
1. no CPU assigned tasks in your queue/cache; and
2. GPU stops processing tasks when the computer is in use.
Is this correct?

Without knowing more about your settings and without being able to see your computers (since they are hidden), it is difficult to know what is causing the situation.

For your hidden computers, you could make them visible by going to your: SETI@home preferences and then checking the box for: Should SETI@home show your computers on its web site?
Another option is to simply send in your reply the Computer_ID that is visible in the left column on the webpage for Your computers.

FYI, there are two types of settings: web-based and local. If you have local settings, they override your web-based settings.
But there are no

Cheers,
Rob :-}
ID: 1799417 · Report as offensive
AMDave
Volunteer tester

Send message
Joined: 9 Mar 01
Posts: 234
Credit: 11,671,730
RAC: 0
United States
Message 1799424 - Posted: 29 Jun 2016, 18:42:22 UTC - in response to Message 1799295.  
Last modified: 29 Jun 2016, 18:50:20 UTC

I will assume that you don’t want the GPU to process WUs while your computer is in use.

Open the BOINC manager to Advanced View.  Select Options from the menu bar, then Computing Preferences.  On the Computing tab, make sure that Suspend when computer is in use, is not selected.  Under Usage limits enter a percentage of the amount of CPU cores to be used for crunching.  If you made any changes, select OK.  This changes the setting locally (for the current machine running BOINC).

Next, go to your Your account, scroll to Preferences, and select SETI@home preferences.  Select Edit preferences.  Check the Use CPU box, then click the Update preferences button.

You can also modify the number of CPU cores to be used via Your account.  Scroll to Preferences, and select Computing preferences.  On the bottom of the next page, select Edit preferences.  From here, change the number in the % of the CPUs field.  This changes the setting globally (for all your machines running BOINC).

Remember, you will need to allow for 1 CPU core to assist the GPU/WU when the GPU does process WUs.  So, allocate CPU usage accordingly.
ID: 1799424 · Report as offensive
Beamer-1

Send message
Joined: 19 May 99
Posts: 9
Credit: 25,250,186
RAC: 17
Message 1805443 - Posted: 28 Jul 2016, 22:35:00 UTC - in response to Message 1799424.  

Perhaps I'm not being clear.. I would like boinc/seti to Give me more CPU units to chew on.. and fewer GPU units to chew on

% CPUs = 100%
%CPU time = 100%
Store 10 days of work
switch tasks ever 60 Minutes
Request Checkpoint = at most 15 seconds
Use 5G of disc, leave me with 6G of disc
Use no more than 10% of total disc

That's pretty much it

I currently have about 100 GPU tasks holding to be done and about 91 CPU tasks.. which is Way out of balance to what hardware I have.
ID: 1805443 · Report as offensive
Beamer-1

Send message
Joined: 19 May 99
Posts: 9
Credit: 25,250,186
RAC: 17
Message 1805444 - Posted: 28 Jul 2016, 22:39:29 UTC - in response to Message 1799417.  

ID: 8009487 is the beast in question
Situation has been growing for some time not.. at least 6 months. I don't normally check on such things but I noticed recently that Boinc was on pause.. I have GPU set to only run when I'm not on, and the entire Queue was GPU'd.. I "asked" for more units and got some CPU-units and chewed on them for a while..
GPU stopping on use is fine.. but there are currently 100 GPU units and 91 CPU units in Queue which I think is Way out of proportion with my hardwares abilities..
or not.. I don't know.. :)
ID: 1805444 · Report as offensive
Profile Stubbles
Volunteer tester
Avatar

Send message
Joined: 29 Nov 99
Posts: 358
Credit: 5,909,255
RAC: 0
Canada
Message 1805497 - Posted: 29 Jul 2016, 2:42:55 UTC - in response to Message 1805443.  

Perhaps I'm not being clear.. I would like boinc/seti to Give me more CPU units to chew on.. and fewer GPU units to chew on
...
Store 10 days of work
...
I currently have about 100 GPU tasks holding to be done and about 91 CPU tasks.. which is Way out of balance to what hardware I have.

The max # of task per device is 100, so "Store 10 days of work" gives you that for each since both could crunch 100 within 10 days.
If you had 2 CPUs and 2 GPUs, you'd get:
100 tasks for CPUs (in total) since the device is the motherboard; and
200 tasks for GPUs (2 * 100)

If you diminish the value for: "Store 10 days of work" (I suggest doing -1 per day),
you should reach a point where you have less than 100 tasks waiting for the GPU
and still have the max of 100 for the CPU.
Continue to do this until you have less than 100 for the CPU and revert back to the previous value.

From my perspective though, having 100 tasks waiting for the GPU is not an issue for S@h since the expiry date of the task is usually 2+ months in the future. If your computer was disconnected from the internet for a few days, having the maximum of 100 per device would allow your rig to keep on crunching until the queues became empty.

Hope this helps a bit more.
Feel free to ask for clarifications or Qs if needed,
Rob :-)
ID: 1805497 · Report as offensive
Profile Stubbles
Volunteer tester
Avatar

Send message
Joined: 29 Nov 99
Posts: 358
Credit: 5,909,255
RAC: 0
Canada
Message 1805501 - Posted: 29 Jul 2016, 2:56:30 UTC - in response to Message 1805444.  

ID: 8009487 is the beast in question

I looked at your PC#8009487 and it has many tasks that didn't complete.
All of them (except for the last 3) reported ending abruptly with:
Error while computing
either directly at the beginning of the task or midway through.
And all of those (except for 1) were processed by the CPU.

In my 3 months here, I have only come across this when GPUs are overclocked too high or when there is a heat issue.
I'm guessing the same could be happening on your CPU...but I'll let others jump in with possible better explanations...as the
CPU time (sec) = 0.00
could possibly be from a faulty HDD or RAM.
ID: 1805501 · Report as offensive
Profile BilBg
Volunteer tester
Avatar

Send message
Joined: 27 May 07
Posts: 3720
Credit: 9,385,827
RAC: 0
Bulgaria
Message 1805509 - Posted: 29 Jul 2016, 3:48:43 UTC - in response to Message 1805501.  

ID: 8009487 is the beast in question

I looked at your PC#8009487 and it has many tasks that didn't complete.
All of them (except for the last 3) reported ending abruptly with:
Error while computing
either directly at the beginning of the task or midway through.
And all of those (except for 1) were processed by the CPU.

For:
Exit status 3221226091 (0xc000026b) Unknown exit code
(unknown error) - exit code -1073741205 (0xc000026b)

Read:
"1073741205 Error Code (Unknown Error)"
https://setiathome.berkeley.edu/forum_thread.php?id=79695

Advice for now:
"Exit BOINC manually before restarting Windows"
http://setiathome.berkeley.edu/forum_thread.php?id=79855&postid=1800530#1800530


P.S.
Technical note:
If you convert Hex C000026B to unsigned it gives Decimal 3221226091
If you convert Hex C000026B to signed it gives Decimal -1073741205
So (if using 4 bytes):
3221226091 == -1073741205 == 0xc000026b

The poster ("James W") incorrectly named his thread:
"1073741205 Error Code (Unknown Error)"

As you see it should be:
"-1073741205 Error Code (Unknown Error)"
 
 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 
ID: 1805509 · Report as offensive
Beamer-1

Send message
Joined: 19 May 99
Posts: 9
Credit: 25,250,186
RAC: 17
Message 1805731 - Posted: 30 Jul 2016, 0:38:50 UTC - in response to Message 1805497.  

OK.. but what need I do to get a better balance of 8cpu : 1gpu?
(I'll try what you suggest re clearing the backload)

Looking a bit deeper I notice that the GPU units are significantly shorter so perhaps this is much ado about nothing..
ID: 1805731 · Report as offensive
Beamer-1

Send message
Joined: 19 May 99
Posts: 9
Credit: 25,250,186
RAC: 17
Message 1805733 - Posted: 30 Jul 2016, 0:44:12 UTC - in response to Message 1805501.  

I hadn't noticed any error messages..
Zero over clocking but in May I had a crash and there were a number of restarts (dozens and dozens.. long story) I may have dropped a bunch in all that
I will take the advice of stop before reboot where ever possible.
ID: 1805733 · Report as offensive
Beamer-1

Send message
Joined: 19 May 99
Posts: 9
Credit: 25,250,186
RAC: 17
Message 1805735 - Posted: 30 Jul 2016, 0:46:55 UTC - in response to Message 1805733.  

OK looked at the error messages for July 29.. don't know why that happened but I only had 2 cpu units when I checked last.. the exact opposite of what I was seeking (should have kept my mouth shut)
ID: 1805735 · Report as offensive
Profile BilBg
Volunteer tester
Avatar

Send message
Joined: 27 May 07
Posts: 3720
Credit: 9,385,827
RAC: 0
Bulgaria
Message 1805800 - Posted: 30 Jul 2016, 10:15:41 UTC - in response to Message 1805731.  
Last modified: 30 Jul 2016, 10:31:56 UTC

OK.. but what need I do to get a better balance of 8cpu : 1gpu?

If you want "Store 10 days of work" the SETI@home server just can't obey to that (other projects can) unless you have very slow computer (e.g. made 10-15 years ago)
Because there is a limit on the number of tasks (it's not only the time limit you set in "days of work")
So the SETI@home server will just fill CPU and GPU to the max possible of 100 + 100 tasks
(Your BOINC will continue to ask for more because 100 + 100 tasks are probably less than "10 days of work" but the SETI@home server will say "No" - visible in Event Log (Ctrl+Shift+E))

But if you have error tasks on CPU the server will restrict even more sending tasks "for CPU" (reduce the #) until some CPU tasks are finished and validated OK

If you set for "1 days of work" the server may be able to "balance"
(it may be set even to fraction e.g. "0.1 days of work")

Look at your computer page and see how the SETI@home server knows how much time every device CPU / GPU is allowed to compute
"While BOINC is running, fraction of time computing is allowed"
"While is BOINC running, fraction of time GPU computing is allowed"

So if you really (in fact) allow less time for GPU to compute the server will "balance" for that.


Looking a bit deeper I notice that the GPU units are significantly shorter

"GPU units" - there is no such thing, they are the same and sent for CPU / GPU / ARM, Windows / Linux / Android ...
You get those WUs "for GPU" but the same WU will be sent to some other computer "for CPU"
They are not "shorter", just your GPU can compute them "significantly" faster than one core of your CPU
 
 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 
ID: 1805800 · Report as offensive

Questions and Answers : GPU applications : Fewer GPU units please


 
©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.