| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Completed Research
Forum: Help Conquer Cancer
Thread: HCC GPU APP HAS LAUNCHED. |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 363
|
|
| Author |
|
|
nanoprobe
Master Cruncher Classified Joined: Aug 29, 2008 Post Count: 2998 Status: Offline Project Badges: 
|
What does the... <avg_ncpus>1.0</avg_ncpus> That is set to say how much of a CPU can be used. 1=use the whole core. .5= use 1/2 a core. .33=use 1/3 or a core etc. Since WCG recommends an entire core for each task being run I use 1 there. <max_ncpus>0.783972</max_ncpus> How many cores do you want to use Why 0.783972 ? Why not, say, 0.75 ? Or 1.25 for the 'max_ncpus' ? Does the machine need to read that much precision (6-decimal places)? Have no idea. Only app_info I've seen that parameter in is for Nvidia. I've always used whole numbers but I've never gone above 2. (That was for POEM not WCG so don't get confused.) 1 is fine for WCG
In 1969 I took an oath to defend and protect the U S Constitution against all enemies, both foreign and Domestic. There was no expiration date.
  |
||
|
|
Bearcat
Master Cruncher USA Joined: Jan 6, 2007 Post Count: 2803 Status: Offline Project Badges: 
|
What does the... <avg_ncpus>1.0</avg_ncpus> <max_ncpus>0.783972</max_ncpus> ... suppose to mean (inside the particular context it is in)? Why 0.783972 ? Why not, say, 0.75 ? Or 1.25 for the 'max_ncpus' ? Does the machine need to read that much precision (6-decimal places)? ; That max_ncpus looks like what's setting the power used in our GPU's to use. Tempting to adjust that to see what happens.
Crunching for humanity since 2007!
 |
||
|
|
nanoprobe
Master Cruncher Classified Joined: Aug 29, 2008 Post Count: 2998 Status: Offline Project Badges:
|
What does the... <avg_ncpus>1.0</avg_ncpus> <max_ncpus>0.783972</max_ncpus> ... suppose to mean (inside the particular context it is in)? Why 0.783972 ? Why not, say, 0.75 ? Or 1.25 for the 'max_ncpus' ? Does the machine need to read that much precision (6-decimal places)? ; That max_ncpus looks like what's setting the power used in our GPU's to use. Tempting to adjust that to see what happens. It will probably slow you down because you're now dedicating 2 CPU cores per 1 GPU task.
In 1969 I took an oath to defend and protect the U S Constitution against all enemies, both foreign and Domestic. There was no expiration date.
|
||
|
|
Bearcat
Master Cruncher USA Joined: Jan 6, 2007 Post Count: 2803 Status: Offline Project Badges:
|
Probably. Would like to know if there is something in the cc config file that could dictate core power to use but suppose its in the wu code.
----------------------------------------Tonights update show 433,737 crunching 4 wu's since last update. Will be trying 6 in the morning.
Crunching for humanity since 2007!
|
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
The max _ncpus can be left at 1, it's the avg_ncpus that tells BOINC Manager how much CPU to allot to a task. I'm running 8 tasks on a 7870 with an FX-4100. To get this, I set avg_ncpus to 0.5 to get 8 instances on 4 cores. It's been running smooth for more than 48 hrs now with no errors or invalids.
|
||
|
|
Bearcat
Master Cruncher USA Joined: Jan 6, 2007 Post Count: 2803 Status: Offline Project Badges:
|
The max _ncpus can be left at 1, it's the avg_ncpus that tells BOINC Manager how much CPU to allot to a task. I'm running 8 tasks on a 7870 with an FX-4100. To get this, I set avg_ncpus to 0.5 to get 8 instances on 4 cores. It's been running smooth for more than 48 hrs now with no errors or invalids. Can you post crunching times doing 8 along with temps and fan speed? I have a 7870 ghz edition and crunching 4 would complete around 3 min. Just raised it to 6 and my times jumped to around 4:30.
Crunching for humanity since 2007!
|
||
|
|
dskagcommunity
Senior Cruncher Austria Joined: May 10, 2011 Post Count: 219 Status: Offline Project Badges: 
|
The max _ncpus can be left at 1, it's the avg_ncpus that tells BOINC Manager how much CPU to allot to a task. I'm running 8 tasks on a 7870 with an FX-4100. To get this, I set avg_ncpus to 0.5 to get 8 instances on 4 cores. It's been running smooth for more than 48 hrs now with no errors or invalids. Can you post crunching times doing 8 along with temps and fan speed? I have a 7870 ghz edition and crunching 4 would complete around 3 min. Just raised it to 6 and my times jumped to around 4:30. What CPU you have? 3 mins is high (i get that too with oolldd quad cpu and 7950, and i know im hardly stuck on low CPU power). dont think it will help more to raise the instances. ---------------------------------------- [Edit 1 times, last edit by dskagcommunity at Oct 30, 2012 4:23:51 PM] |
||
|
|
pirogue
Veteran Cruncher USA Joined: Dec 8, 2008 Post Count: 685 Status: Offline Project Badges: 
|
The max _ncpus can be left at 1, it's the avg_ncpus that tells BOINC Manager how much CPU to allot to a task. I'm running 8 tasks on a 7870 with an FX-4100. To get this, I set avg_ncpus to 0.5 to get 8 instances on 4 cores. It's been running smooth for more than 48 hrs now with no errors or invalids. Can you post crunching times doing 8 along with temps and fan speed? I have a 7870 ghz edition and crunching 4 would complete around 3 min. Just raised it to 6 and my times jumped to around 4:30. (60s * 3) / 4 = 45s (60s * 4.5) / 6 = 45s You're still doing 1 WU every 45 seconds. That seems a little slow even at 4. Are you running something besides HCC GPU? On the plus side, that's still 400K+ WCG PPD.  I can't tell if you're actually using the <*_ncpu> settings, but if I were you, I'd stick to changing the <count> setting, not messing with the <*_ncpu> settings. [Edit 1 times, last edit by pirogue at Oct 30, 2012 4:57:36 PM] |
||
|
|
JacobKlein
Cruncher Joined: Aug 21, 2007 Post Count: 21 Status: Offline |
Robert,
----------------------------------------Here is what I have found out regarding our problem: Per the WCG Getting Started "Graphics Card Processing" section, here http://www.worldcommunitygrid.org/help/viewTopic.do?shortName=GPU , the question "What if my computer stutters while World Community Grid is utilizing my graphics card?" describes one possible workaround. Essentially, you can configure BOINC to not utilize the GPU while the user is active. This can be configured at WCG (if your BOINC client is using Web profile settings), or in BOINC directly (Tools -> Computing preferences -> Processing Usage -> Uncheck "Use GPU while computer is in use") Another workaround is to disable WCG HCC GPU processing on your main GPU that's connected to your display, using <exclude_gpu> configuration in the cc_config.xml file. That's under the "If you have more than one graphics card, can you limit which graphics card is used?" question in that "Graphics Card Processing" article. I pressed further to see if WCG could improve their GPU application to run better while the user is active. I still hope they find a way, but I've been told it is not possible, per knreeds's post here - http://www.worldcommunitygrid.org/forums/wcg/viewpostinthread?post=395454 ... So we cannot rely on the situation improving from an application standpoint. I personally don't want GPU operations suspended entirely while I use the computer (10 hours a day). After all, other projects' tasks (Poem@Home, and most from GPUGrid.Net) process without interfering while I work, and they should get to run! So, my workaround (listed here http://www.worldcommunitygrid.org/forums/wcg/viewpostinthread?post=396551 ) is to manually suspend (using BOINC's Advanced View, Tasks Tab, All Tasks, sorted by Application) all WCG HCC GPU tasks when I sit down at the PC (so the GPU gets to work on other projects' tasks), then unsuspend those tasks when I get up from my PC. It's babysitting, and I don't like it, but for maximum non-interfering processing, it is my only workaround. Kind regards, Jacob Klein I've set my two desktops to No New Tasks for World Community Grid, and I find that it makes both of them much more responsive to the keyboard and the mouse. Since it would otherwise be running HCC GPU workunits most of the time and seldom any WCG CPU workunits, it looks like the HCC GPU workunits are rather poor at sharing some computer resource with programs run by the user rather than started automatically. My guess is that the poorly shared resource is access to the graphics board. Could you investigate, and try to add a way to prevent this? Perhaps add a user setting for each graphics board for what fraction of its resources the HCC GPU workunits are allowed to use? I suspect that users with graphics boards that have no monitors attached would want to allow the workunits to use those entire graphics boards, but those with monitors attached would want to limit this use. Another idea - allow users to make server settings so that GPU workunits can only be sent during certain hours of the day and night, but CPU workunits can be sent any time. Another idea to help those with poor keyboard and mouse response: Provide a way to tell the BOINC client not to start any more HCC GPU workunits. It's less important whether this also suspends any such workunits now running. No need to shut down the GPU entirely - several other BOINC projects offer GPU workunits with much less slowdown of the keyboard and mouse response [Edit 2 times, last edit by JacobKlein at Oct 30, 2012 7:14:59 PM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
The max _ncpus can be left at 1, it's the avg_ncpus that tells BOINC Manager how much CPU to allot to a task. I'm running 8 tasks on a 7870 with an FX-4100. To get this, I set avg_ncpus to 0.5 to get 8 instances on 4 cores. It's been running smooth for more than 48 hrs now with no errors or invalids. Can you post crunching times doing 8 along with temps and fan speed? I have a 7870 ghz edition and crunching 4 would complete around 3 min. Just raised it to 6 and my times jumped to around 4:30. With 8 running my times went up to about 6:30 or so. Have now brought it down to 6 at a time and seeing 4:20. I'm really just playing around to see what it can handle without creating errors. From what I was told, memory speed is a big factor on times. I have 1600 ram, but was told 1800 is minimum to keep things going as fast as they can. My card is the Sapphire GHz edition ( http://www.sapphiretech.com/presentation/prod...;psn=&lid=1&leg=0 ) with the fan on auto (never goes over 30%) it is sitting just above 50C. |
||
|
|
|