It's been a while since I changed the utilization factor but if I remember correctly the new setting applies to all tasks in cache but the displayed info is wrong until you start running tasks downloaded after the setting is changed.
As you've verified that the setting is in effect and you're running x2 on the GPU it's just the displayed info that's wrong but that will sort itself out as you run through you cached work so just let it run.
... my Einstein pref is set at "GPU utilization factor of BRP apps 0.5"
You posted this message at around 1.00AM UTC on 2nd Nov. Any GPU tasks reported before that time would have been crunched 1x. You have two such tasks (BRP5) still showing in your tasks list, and these would have been crunched 1x. They have average values for elapsed time/cpu time of around 57ksecs/2.2ksecs. For your purposes, you want to look at run time and make sure that when running 2x, the run time of a task is less than double 57Ksecs, ie less than 114ksecs.
You have a couple of BRP5 tasks that have been completed (but not yet validated) that have run times around 106ksecs so it appears that you can gain a little by running 2x. However, the worrying thing is that you also have several tasks that have errored out, a pair after significant crunching time had already elapsed and the rest almost immediately after commencing. Either something happened to your machine that caused a bunch of tasks to fail or perhaps the stress of crunching was too great for your card to handle. If you continue to get errors like this you should change your GPU utilization pref back so as to crunch these tasks 1x. The benefit from 2x is marginal and is completely negated if tasks are going to error out repeatedly.
I've addressed some issues raised by Gary's last post - my comments are in boldface.
"... my Einstein pref is set at "GPU utilization factor of BRP apps 0.5"
You posted this message at around 1.00AM UTC on 2nd Nov. Any GPU tasks reported before that time would have been crunched 1x. You have two such tasks (BRP5) still showing in your tasks list, and these would have been crunched 1x. They have average values for elapsed time/cpu time of around 57ksecs/2.2ksecs. For your purposes, you want to look at run time and make sure that when running 2x, the run time of a task is less than double 57Ksecs, ie less than 114ksecs. Shouldn't the "estimated computation size (ECS)" of each task also be taken into consideration as well? For instance so far I've noticed that many of my Einstein GPU tasks have an ECS of 450,000 GFLOPs, although the 2 running tasks now are both 280,000 GFLOPs. ]
You have a couple of BRP5 tasks that have been completed (but not yet validated) that have run times around 106ksecs so it appears that you can gain a little by running 2x. However, the worrying thing is that you also have several tasks that have errored out, a pair after significant crunching time had already elapsed and the rest almost immediately after commencing. Either something happened to your machine that caused a bunch of tasks to fail or perhaps the stress of crunching was too great for your card to handle. If you continue to get errors like this you should change your GPU utilization pref back so as to crunch these tasks 1x. The benefit from 2x is marginal and is completely negated if tasks are going to error out repeatedly. I think "Code 18" (see http://www.urbandictionary.com/define.php?term=Code%2018) applies to the tasks that "erred out" almost immediately after commencing. I can't recall why the other two tasks with significant crunching time failed to complete, but I'll start keeping written records if anymore errors occur so I'll know for sure what caused them.
I can't recall why the other two tasks with significant crunching time failed to complete, but I'll start keeping written records if anymore errors occur so I'll know for sure what caused them. [/b]
No need to keep written records. BOINC does that for you.
It will show in the Event Log (CTRL+SHIFT+E) what the error was prior to upload.
It will show you in your Tasks list, under the Error option (top and bottom of the screen). In this case these are your erroneous tasks.
Now you can click on each of the Tasks' IDs (task names) to get to the bit of information that BOINC sent along to what kind of error it came across.
Example given http://einsteinathome.org/task/408774290 will show:
This error translates into the Microsoft error "The application failed to initialize properly (0xc0000135)". Or according to the BOINC developers, it's "Usually it is because some library that is being linked to the executable has a different version of the C Runtime Library included in its manifest."
Not something you can fix. So also not something you have to worry about. Just report such errors on the forums and someone will look into it.
Thanks Ageless I never really thought about the errors very much since I usually don't get a lot, although now that I'm writing this I have no idea what a "lot" is ie 2%? 5%? 10% or more of all tasks? A few times I did notice problems on my end that I presumed caused tasks ending in errors so I never thought about some errors not being my fault (or my PC's fault) I must have a guilty conscience huh?
I started using the snip tool to keep records because I noticed that BOINC only seems to display records of about 20 or so tasks - completed & pending. I've already started looking at the tasks data here but had no idea where to find any error messages b4 reading your last message.
My latest brainstorm has been to concentrate on Einstein tasks exclusively for a wk or 2 to get more results ASAP and have some sort of DB to compare with my results running one GPU task at a time
UPDATE:
As of 12-11-13 I haven't had anymore errors for any GPU tasks after the 2 noted by Gary in Message 127798 - Posted 5 Nov 2013. I haven't noticed any other errors either and I look at the BOINC manager at least one daily.
My active projects now are Einstein@Hm (2 GPU tasks running simultaneously); Rosetta@Hm and LHC Test4Theory@Hm these projects run 1 task each so my PC is running a total of 4 tasks at once with these settings: Use 3/4 of CPU Cores
Use up to 100% of CPU Time (I presume that applies to the 3 cores in use) Use up to 60% of memory while PC in use
Use up to 90% of memory while PC is idle Use up to 75% of page file (= 6 GB out of 8 Total)
Leave apps in memory while suspended - although I think this is a moot point(?) since my PC runs BOINC 24/7
A person who never made a mistake never tried anything new.Albert Einstein
... As of 12-11-13 I haven't had anymore errors ...
Good to hear that things seem to be running smoothly.
Quote:
Leave apps in memory while suspended - although I think this is a moot point(?) since my PC runs BOINC 24/7
If you have enough 'memory' (physical RAM + swap space) it's a good idea to use this setting. Most tasks do save a checkpoint reasonably frequently but some don't. If a task (with infrequent or no checkpointing) gets suspended whilst crunching, you stand to lose a significant amount of crunch time - back to whenever the last checkpoint (if any) was saved. Keeping the task image in memory will avoid this loss when the task resumes. If the image isn't kept in memory, the last checkpoint will have to be reloaded and crunching will resume from there.
Of course, there will only be an image in memory if BOINC wasn't stopped subsequent to the task being suspended. When BOINC stops and restarts, the last checkpoint must be used for all tasks in progress.
Last 2013 Update(?): Out of idle curiosity I changed my settings from "use 75% of CPUs" & "Use 100% of CPU time" (presumably on those 3 cores, not the 4th one) to "use all (4/4) CPUs" & "use up to 75% of CPU time" which has resulted in my PC running 5 tasks at once instead of 4, the 5th task is another one for Rosetta@Hm. I only changed these settings a few daze ago but so far my PC's RAM usage doesn't appear to have increased much from what I've seen and the CPU usage does appear somewhat higher but not enuff to intefere with my usual activities ie browsing the web or checking email. I can even watch Netflix while BOINC is active although I usually suspend it because Netflix appears to be herky-jerky trying to get by on whatever little GPU resources it can scrounge up while Einstein@Hm is running GPU projects. MERRY XMAS - HAPPY NEW YEAR everyone!
How could they tell?Dorothy Parker(after being informed ex-President Calvin Coolidge had died)
It's been a while since I
)
It's been a while since I changed the utilization factor but if I remember correctly the new setting applies to all tasks in cache but the displayed info is wrong until you start running tasks downloaded after the setting is changed.
As you've verified that the setting is in effect and you're running x2 on the GPU it's just the displayed info that's wrong but that will sort itself out as you run through you cached work so just let it run.
RE: ... my Einstein pref is
)
You posted this message at around 1.00AM UTC on 2nd Nov. Any GPU tasks reported before that time would have been crunched 1x. You have two such tasks (BRP5) still showing in your tasks list, and these would have been crunched 1x. They have average values for elapsed time/cpu time of around 57ksecs/2.2ksecs. For your purposes, you want to look at run time and make sure that when running 2x, the run time of a task is less than double 57Ksecs, ie less than 114ksecs.
You have a couple of BRP5 tasks that have been completed (but not yet validated) that have run times around 106ksecs so it appears that you can gain a little by running 2x. However, the worrying thing is that you also have several tasks that have errored out, a pair after significant crunching time had already elapsed and the rest almost immediately after commencing. Either something happened to your machine that caused a bunch of tasks to fail or perhaps the stress of crunching was too great for your card to handle. If you continue to get errors like this you should change your GPU utilization pref back so as to crunch these tasks 1x. The benefit from 2x is marginal and is completely negated if tasks are going to error out repeatedly.
Cheers,
Gary.
RE: I've addressed some
)
You posted this message at around 1.00AM UTC on 2nd Nov. Any GPU tasks reported before that time would have been crunched 1x. You have two such tasks (BRP5) still showing in your tasks list, and these would have been crunched 1x. They have average values for elapsed time/cpu time of around 57ksecs/2.2ksecs. For your purposes, you want to look at run time and make sure that when running 2x, the run time of a task is less than double 57Ksecs, ie less than 114ksecs. Shouldn't the "estimated computation size (ECS)" of each task also be taken into consideration as well? For instance so far I've noticed that many of my Einstein GPU tasks have an ECS of 450,000 GFLOPs, although the 2 running tasks now are both 280,000 GFLOPs. ]
You have a couple of BRP5 tasks that have been completed (but not yet validated) that have run times around 106ksecs so it appears that you can gain a little by running 2x. However, the worrying thing is that you also have several tasks that have errored out, a pair after significant crunching time had already elapsed and the rest almost immediately after commencing. Either something happened to your machine that caused a bunch of tasks to fail or perhaps the stress of crunching was too great for your card to handle. If you continue to get errors like this you should change your GPU utilization pref back so as to crunch these tasks 1x. The benefit from 2x is marginal and is completely negated if tasks are going to error out repeatedly. I think "Code 18" (see http://www.urbandictionary.com/define.php?term=Code%2018) applies to the tasks that "erred out" almost immediately after commencing. I can't recall why the other two tasks with significant crunching time failed to complete, but I'll start keeping written records if anymore errors occur so I'll know for sure what caused them.
CIAO - Auf Wiedersehen - ALOHA - Saludos - SHALOM - Sayonara - ADIEU - 明朝会ADIOS - להתראות - G'DAY!
RE: I can't recall why the
)
No need to keep written records. BOINC does that for you.
It will show in the Event Log (CTRL+SHIFT+E) what the error was prior to upload.
It will show you in your Tasks list, under the Error option (top and bottom of the screen). In this case these are your erroneous tasks.
Now you can click on each of the Tasks' IDs (task names) to get to the bit of information that BOINC sent along to what kind of error it came across.
Example given http://einsteinathome.org/task/408774290 will show:
(unknown error) - exit code -1073741515 (0xc0000135)
]]>
Something that the next erroneous task also shows.
http://einsteinathome.org/task/408753881 says:
(unknown error) - exit code -1073741515 (0xc0000135)
Activated exception handling...
This error translates into the Microsoft error "The application failed to initialize properly (0xc0000135)". Or according to the BOINC developers, it's "Usually it is because some library that is being linked to the executable has a different version of the C Runtime Library included in its manifest."
Not something you can fix. So also not something you have to worry about. Just report such errors on the forums and someone will look into it.
Thanks Ageless I never really
)
Thanks Ageless I never really thought about the errors very much since I usually don't get a lot, although now that I'm writing this I have no idea what a "lot" is ie 2%? 5%? 10% or more of all tasks? A few times I did notice problems on my end that I presumed caused tasks ending in errors so I never thought about some errors not being my fault (or my PC's fault) I must have a guilty conscience huh?
I started using the snip tool to keep records because I noticed that BOINC only seems to display records of about 20 or so tasks - completed & pending. I've already started looking at the tasks data here but had no idea where to find any error messages b4 reading your last message.
My latest brainstorm has been to concentrate on Einstein tasks exclusively for a wk or 2 to get more results ASAP and have some sort of DB to compare with my results running one GPU task at a time
CIAO - Auf Wiedersehen - ALOHA - Saludos - SHALOM - Sayonara - ADIEU - 明朝会ADIOS - להתראות - G'DAY!
UPDATE: As of 12-11-13 I
)
UPDATE:
As of 12-11-13 I haven't had anymore errors for any GPU tasks after the 2 noted by Gary in Message 127798 - Posted 5 Nov 2013. I haven't noticed any other errors either and I look at the BOINC manager at least one daily.
My active projects now are Einstein@Hm (2 GPU tasks running simultaneously); Rosetta@Hm and LHC Test4Theory@Hm these projects run 1 task each so my PC is running a total of 4 tasks at once with these settings:
Use 3/4 of CPU Cores
Use up to 100% of CPU Time (I presume that applies to the 3 cores in use)
Use up to 60% of memory while PC in use
Use up to 90% of memory while PC is idle
Use up to 75% of page file (= 6 GB out of 8 Total)
Leave apps in memory while suspended - although I think this is a moot point(?) since my PC runs BOINC 24/7
A person who never made a mistake never tried anything new. Albert Einstein
CIAO - Auf Wiedersehen - ALOHA - Saludos - SHALOM - Sayonara - ADIEU - 明朝会ADIOS - להתראות - G'DAY!
RE: ... As of 12-11-13 I
)
Good to hear that things seem to be running smoothly.
If you have enough 'memory' (physical RAM + swap space) it's a good idea to use this setting. Most tasks do save a checkpoint reasonably frequently but some don't. If a task (with infrequent or no checkpointing) gets suspended whilst crunching, you stand to lose a significant amount of crunch time - back to whenever the last checkpoint (if any) was saved. Keeping the task image in memory will avoid this loss when the task resumes. If the image isn't kept in memory, the last checkpoint will have to be reloaded and crunching will resume from there.
Of course, there will only be an image in memory if BOINC wasn't stopped subsequent to the task being suspended. When BOINC stops and restarts, the last checkpoint must be used for all tasks in progress.
Cheers,
Gary.
Last 2013 Update(?): Out of
)
Last 2013 Update(?): Out of idle curiosity I changed my settings from "use 75% of CPUs" & "Use 100% of CPU time" (presumably on those 3 cores, not the 4th one) to "use all (4/4) CPUs" & "use up to 75% of CPU time" which has resulted in my PC running 5 tasks at once instead of 4, the 5th task is another one for Rosetta@Hm. I only changed these settings a few daze ago but so far my PC's RAM usage doesn't appear to have increased much from what I've seen and the CPU usage does appear somewhat higher but not enuff to intefere with my usual activities ie browsing the web or checking email. I can even watch Netflix while BOINC is active although I usually suspend it because Netflix appears to be herky-jerky trying to get by on whatever little GPU resources it can scrounge up while Einstein@Hm is running GPU projects. MERRY XMAS - HAPPY NEW YEAR everyone!
How could they tell? Dorothy Parker(after being informed ex-President Calvin Coolidge had died)
CIAO - Auf Wiedersehen - ALOHA - Saludos - SHALOM - Sayonara - ADIEU - 明朝会ADIOS - להתראות - G'DAY!