| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Completed Research
Forum: Help Conquer Cancer
Thread: GPU Optimisations |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 198
|
|
| Author |
|
|
branjo
Master Cruncher Slovakia Joined: Jun 29, 2012 Post Count: 1892 Status: Offline Project Badges: 
|
I also tried branjo's option of 24 concurrent tasks on my not overclocked 7770 GPU and all went well without errors or restarts. I did 4 runs of 24 each with all 8 threads of my not overclocked i7-2600 dedicated to the GPU-tasks. The first run I started 24 tasks at once, but after each run the GPU- and CPU-part of the tasks separates more and more, leading to lower elapsed times. Here the 4 runs with mean total run time for 24 tasks and runtime per task: 58:29 - 02:26 55:28 - 02:19 54:28 - 02:16 53:47 - 02:14 Hey C.P. Can get good comparison between the non OC'd and the OC'd 7770 as I'm also running a 2600 cpu that isn't OC'd. Same app_config as you I think - .333 cpu and .041 gpu per task for 24 concurrent GPU tasks. Task average finish time is 46 mins for mine. I can't wait for Rob's math on that, the ppd seems ridiculous for a cheap graphics card. Rob - you sure that math in your last post was correct ?? (I got no idea myself) ARH - it was too high then (didn't refresh my browsewr before I posted - sorry) You can bet your new GPU on it  With 46 mins per WU and 24 concurrent, your production should be approximately 341K WCG points (48.8K BOINC Credits) per day  Cheers   Crunching@Home since January 13 2000. Shrubbing@Home since January 5 2006  |
||
|
|
Crystal Pellet
Veteran Cruncher Joined: May 21, 2008 Post Count: 1404 Status: Offline Project Badges: 
|
Looks again... 26.77 results per hour times 24 hours in a day times 65 credit per result times 7 points per credit... don't know what your abacus makes of this or which multiplicator is incorrect. Sorry Rob, misunderstood you. Didn't realize you was still living in the WCG points age  |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Looks again... 26.77 results per hour times 24 hours in a day times 65 credit per result times 7 points per credit... don't know what your abacus makes of this or which multiplicator is incorrect. Sorry Rob, misunderstood you. Didn't realize you was still living in the WCG points age Don't know what you think ppd or PPD stands for, but in your world it would then have to be cpd or CPD, so think I was pretty specific [in living in WCG's world ;P] P.S. The SNURK sig does also give ... ppd. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
@ underwater, in the time it took to take out one card and replace it, I am sure you could have pulled in GPU units. I suggest putting your AMD card back in and doing a Project Update. (In Boinc Manager, this is done by going to Projects tab, highlighting World Community Grid line and hitting the Update button on the left. You can then check the result in the Event Log window. keep repeating this until you get some GPU units. You shouldn't have to do this too many times). There is some debate about the optimal amount of units you can process simultaneously, and this would also depend on your CPU. With the same card in one of my machines and an i7 920, I am doing 3 GPU tasks simultaneously. Good luck with yours! I discovered why I was getting the no ATI task message, after new install. When first installing BOINC software and allowing to run through the initial download stage inc tasks (I press no new tasks asap) so required files are in place for app_config to work. I then abort all transfers asap. I then abort all running tasks. This is done so I can configure the app_config file asap. So after shutting down the BOINC client and configuring the app_config file I start up the client again. This is when the no ati work message appeared. Looking at "projects / wcg/ properties" a pop up box appears with info that has "work deferred times for both cpu and gpu" I can only imagine the server create that deferred fetch time by the aborted tasks that were submitted. After deferred time had elapsed - download of WU's commenced. Project update would not clear/reset the deferred time as suggested to try. This info may help others who experience same message, check project properties  |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Reads like a Big Whopper config... Double Whopper....  My current app_config for 7970 and dual 5650's on SR-2 board. 30 concurrent gpu - 29 @ .034482 cpu per gpu @ .793103 gets 1 cpu task 29 gpu task concurrent each gpu task finshing in 24 mins (average) gives 77 completed tasks per hour OR 1 completed gpu task every 47 seconds   [Edit 1 times, last edit by Former Member at Feb 16, 2013 6:59:35 AM] |
||
|
|
coolstream
Senior Cruncher SCOTLAND Joined: Nov 8, 2005 Post Count: 475 Status: Offline Project Badges: 
|
@ underwater, in the time it took to take out one card and replace it, I am sure you could have pulled in GPU units. I suggest putting your AMD card back in and doing a Project Update. (In Boinc Manager, this is done by going to Projects tab, highlighting World Community Grid line and hitting the Update button on the left. You can then check the result in the Event Log window. keep repeating this until you get some GPU units. You shouldn't have to do this too many times). There is some debate about the optimal amount of units you can process simultaneously, and this would also depend on your CPU. With the same card in one of my machines and an i7 920, I am doing 3 GPU tasks simultaneously. Good luck with yours! I discovered why I was getting the no ATI task message, after new install. When first installing BOINC software and allowing to run through the initial download stage inc tasks (I press no new tasks asap) so required files are in place for app_config to work. I then abort all transfers asap. I then abort all running tasks. This is done so I can configure the app_config file asap. So after shutting down the BOINC client and configuring the app_config file I start up the client again. This is when the no ati work message appeared. Looking at "projects / wcg/ properties" a pop up box appears with info that has "work deferred times for both cpu and gpu" I can only imagine the server create that deferred fetch time by the aborted tasks that were submitted. After deferred time had elapsed - download of WU's commenced. Project update would not clear/reset the deferred time as suggested to try. This info may help others who experience same message, check project properties OK, I'm still using app_info, so my method may not work the same with app_config. The main difference between my and your method is that we both start the same way, but when I see that at least one GPU has been downloaded, I then stop BOINC and insert my app_config file and then restart BOINC. I don't cancel anything and just let BOINC replace the running CPU tasks with new GPU tasks when they are automatically downloaded. From your description, it seems to me that you were downloading tasks and then aborting them so I'm amazed that GPU tasks ever started to download after your restart of BOINC. For a more detailed description, see my posts (commencing here from earlier in this thread).  Crunching in memory of my Mum PEGGY, cousin ROPPA and Aunt AUDREY. |
||
|
|
coolstream
Senior Cruncher SCOTLAND Joined: Nov 8, 2005 Post Count: 475 Status: Offline Project Badges:
|
Reads like a Big Whopper config... Double Whopper.... My current app_config for 7970 and dual 5650's on SR-2 board. 30 concurrent gpu - 29 @ .034482 cpu per gpu @ .793103 gets 1 cpu task 29 gpu task concurrent each gpu task finshing in 24 mins (average) gives 77 completed tasks per hour OR 1 completed gpu task every 47 seconds Interesting. On this hex core i7 with HT and 7970 (vanilla), I am running 12 GPU tasks concurrently using 1 CPU and 0.08 GPU. Units are completing in between 7-9 minutes which equates to approx 90 units per hour. I have looked briefly at points achieved (as a measure of what others are doing and a comparison of how well my machines are faring) and my scores are on average the same. Have you compared your points score with your wingmen, and do you tend to get more or less than your results claim? Crunching in memory of my Mum PEGGY, cousin ROPPA and Aunt AUDREY. [Edit 1 times, last edit by coolstream at Feb 16, 2013 10:06:38 AM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Nobody has not got to do anything special when utilizing/modifying app_config.xml [in possible combination to modifying card control settings in the cc_config.xml, such as setting WCG to not use card 0 or 1 or whatever]. The only thing needed is restart the client and the client scheduler will know what to do. That's a big advantage over app_info.xm **.
----------------------------------------Of course, if you have work on the device that after the config mods would no longer run on any resource [for instance you block all HCC-ATI jobs to run on the only ATI device you have in the host], you best abort that work, or just let it sit to expire at deadline [will get server aborted], but that would till then occupy buffer space. Really don't know if the latest [test] clients would identify such 'never-to-be-run' tasks and auto-aborts them. That's a bridge too far expectation at this time I think. ** Will repeat again, the app_info use will be outlawed at WCG whence a 7.0.xx gets officially adopted/endorsed. Learn to use app_config.xml now or be in cram jam when the announcement hits. [Edit 1 times, last edit by Former Member at Feb 16, 2013 10:13:24 AM] |
||
|
|
coolstream
Senior Cruncher SCOTLAND Joined: Nov 8, 2005 Post Count: 475 Status: Offline Project Badges:
|
OK Rob. Message received loud and clear.
----------------------------------------I lost over a week of processing on one machine after a blue screen crash left it unable to connect to localhost. No matter what I did, I just couldn't get BOINC to work again. In another thread, there was the reccomendation to install BOINC as administrator, but by that time, my patience had run out and I went for a totally clean installation of windows. Now that the machine is back up and running, I will start to look at app_config Crunching in memory of my Mum PEGGY, cousin ROPPA and Aunt AUDREY. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Nobody has not got to do anything special when utilizing/modifying app_config.xml [in possible combination to modifying card control settings in the cc_config.xml, such as setting WCG to not use card 0 or 1 or whatever]. The only thing needed is restart the client and the client scheduler will know what to do. That's a big advantage over app_info.xm **. Ok, so I think I know where meathod in my madness came from. Please correct me if I am wrong. Old "app_info" meathod the cache had to be empty before making the file - right ?? New "app_config" no emptying of cache is required before initial making of the XML file ??? I had assumed (wrongly) that the cache needed to be cleared before making up the app_config file in the first instance. Modifying app_config XML file, just shut down client - that one no problems. Have I got it now OR am I still not getting it ? Cheers |
||
|
|
|