Posts by Philadelphia

\n studio-striking\n
1) Message boards : Number crunching : (2) 5870's (Message 1763)
Posted 16 Dec 2011 by Philadelphia
Post:
Going to clear out my Moo's cashe and move on, sucks, I was expecting about 700k a day :(


Dan it was only 7 hours between your first post and your last post, give us a chance to take some naps and analyze the data guy.

Now you know that Moo is a project that uses BOTH of your gpu's on ONE workunit, right?! This could be what you are seeing the 'stream'ing of the data between the cards so ALL of BOTH cards resources are being used on ONE workunit. When you only have one card no 'stream'ing is needed, so the numbers are zero. Remember too your two cards are NOT identical, one has 1gig of memory while the other card has 2gig of memory, this could be why the workunits are taking so long. You MIGHT be able to limit the resources thru config settings but don't quote me on that.


That is true, they are not identical

I was not aware that Woo used both cards to crunch 1 WU since two of them are running at the same time, or at least in BOINC they show two running at the same time.


Dan I am now back to dual AMD 5770's in one machine and have TWO units running at the same time!!! I have no idea why but that is what it is doing, the cards are IDENTICAL and I have rebooted several times! Each unit says it will take about 60 minutes, which is normal for a 5770. We will see!!


That's how mine started, crunching 2 wu's, it wasn't until I changed from no new tasks to allow it to download again, that it went from crunching 2 wu's to 1 wu.

If you look at the top of this thread you'll see a picture of the two wu's crunching.

Strange for sure.
2) Message boards : Number crunching : (2) 5870's (Message 1707)
Posted 14 Dec 2011 by Philadelphia
Post:
PS - Know how you can tell I'm an old fart? I still call them directories instead of folders :)

[old fart mode on]
It was Microsoft that insisted upon calling the sub-divisions folders when Windows got introduced. Us MSDOS based guys continue to call them directories regardless, just to spite Bill Gates :-)

Get a dos prompt window in Win 7 (type CMD) and you will find that DIR & VER & CD.. still work as do other commands ....

Ah the good old days. EMS386 anyone ?
[old fart mode off].



It's my thread so I guess I can get off topic for just a tiny bit, plus I'm an 'old fart'.

What was the o/s before dos, and nooooooooo looking on the internet. I'm talking about pc's, not fortran or cobol, etc.

One will earn an 'old fart badge' for that one, or, if you're a youngster you're well informed.
3) Message boards : Number crunching : Faulty WUs (Message 1697)
Posted 13 Dec 2011 by Philadelphia
Post:
Yikes, those are huge numbers compared to my bad ones.

I would have thought someone from the project would have shared what the issue is and when to expect a fix; or at least an acknowledgement of an issue and 'we're working on it'.
4) Message boards : Number crunching : Faulty WUs (Message 1688)
Posted 12 Dec 2011 by Philadelphia
Post:
I was experiencing a couple of computation errors on my rig during the last two days. As all of my wingmen are getting the same error, there seems to be something wrong with the corresponding WUs. Here's some examples:
http://moowrap.net/workunit.php?wuid=4382462
http://moowrap.net/workunit.php?wuid=4440517
http://moowrap.net/workunit.php?wuid=4415816
All are showing the same error code:
</stderr_txt>
<message>
upload failure: <file_xfer_error>
  <file_name>dnetc_r72_1323456479_401_768_2_0</file_name>
  <error_code>-131</error_code>
</file_xfer_error>

</message>


Interesting you should mention, I am as well, I assumed it was just me.

http://moowrap.net/workunit.php?wuid=4382112
http://moowrap.net/workunit.php?wuid=4414515
http://moowrap.net/workunit.php?wuid=4400706
http://moowrap.net/workunit.php?wuid=4400704
http://moowrap.net/workunit.php?wuid=4400701
http://moowrap.net/workunit.php?wuid=4400680

I believe I had others, they may have been cleared out by Moo. My a few my wingmen also had errors.
5) Message boards : Number crunching : (2) 5870's (Message 1687)
Posted 12 Dec 2011 by Philadelphia
Post:
Thanks Crunch3r.

Any 'chance' it could be boinc, I'm running 6.10.60 and when I went backwards to install it from whatever the boinc current version is I had to install it a second time so it could fix itself, or something like that.

I also can't get boinc to report tasks, I have to do them manually.


G'Day Philadelphia,

I was having the same problem as you with all work being processed before it would ask for work again.

To get your jobs to report straight away you need to create a "cc_config" file.

Open notepad and type this in

<cc_config>
<options>
<report_results_immediately>1</report_results_immediately>
</options>
</cc_config>

Save as cc_config.xml
Place the file in the Application/BOINC folder

Stop and restart BOINC client and it should report results within seconds of them finishing.

I may of gotten information about your computer incorrect before but I was only taking it off what BOINC said so you can blame BOINC.

Conan


Thanks Conan for the information, it did as you said :)

As an aside, it goes in the program data/boinc directory. Tried the Program Files/Boinc path and it didn't work.

Thanks Conan and everyone else for suggestion & thoughts, great bunch of folks.

Dan/Philly

PS - Know how you can tell I'm an old fart? I still call them directories instead of folders :)
6) Message boards : Number crunching : (2) 5870's (Message 1683)
Posted 11 Dec 2011 by Philadelphia
Post:
Thanks Crunch3r.

Any 'chance' it could be boinc, I'm running 6.10.60 and when I went backwards to install it from whatever the boinc current version is I had to install it a second time so it could fix itself, or something like that.

I also can't get boinc to report tasks, I have to do them manually.
7) Message boards : Number crunching : (2) 5870's (Message 1681)
Posted 11 Dec 2011 by Philadelphia
Post:
I'm going to throw another twist into this twisted situation.

I have been watching boinc and it hit me that boinc was finishing a wu in ~15 minutes but not being reported that way at Moo, sooooo, I sat at the computer for a couple of hours and read the newspaper and watched the wu's crunching, I also looked at a wall clock to verify what boinc was showing, rounding to minutes. The wall clock confirmed what boinc was reporting. I dropped off the seconds on the UTC to figure the minutes run. (edit, that brings up another issue, boinc wont upload the wu's after they run, I have to do it manually)

Having nothing to do this morning , right, I copied about 10 wu's reported by Moo, and low and behold 'something stinks in Denmark' (sorry to anyone in Denmark, just a saying )

Look at the difference between WU's, between the actual UTC times, they agree with the times boinc was showing when I watched them crunching and the wall clock, but the run/cpu times don't.

Could it be a Moo problem? Can I detach from the project and download the Moo program again without screwing things up, or is it even necessary?

8) Message boards : Number crunching : (2) 5870's (Message 1680)
Posted 11 Dec 2011 by Philadelphia
Post:
I see that your computer has 2 GB of main memory.


Actually I have 4 1gb sticks installed.




At 8:28:08 on 12/8/2011 it says "Memory: 2.62 GB physical"

This means Boinc is not seeing all of your 4 gig of memory, some could be bad or not seated all the way or something.

Oh and you asked "How do I dedicate 1 core to the GPU? Currently I'm not crunching and cpu projects, only GPU." Answer is you can't in Boinc except on a unit by unit basis and then only when the unit initially starts. When it finishes and another new one starts you are back to random again. If you babysit the pc you can catch each unit as it starts and assign it to a core thru the Task Manager but this isn't practical! There IS a program that the user says will do this for you automatically, I will look for his message.


Thanks Mike, I'll check the seating of the sticks.

If you look at my last thread, at the bottom, is a Vista pic where Vista see's the 4gb.

Curious, why would Vista see it but not BOINC.

Dan/Philly
9) Message boards : Number crunching : (2) 5870's (Message 1676)
Posted 11 Dec 2011 by Philadelphia
Post:
G'Day Philadelphia,

I have 2x 5870 cards in 2 different machines (so 4 cards all together), and their run times are a lot quicker than yours appear to be.

You have a 5870 and a 5970 (according to BOINC, effectively 3 cards), so your output on a computer basis should trump mine quite easily, yet it doesn't.


Actually I have 2 5870's, one with 1gb of memory and the other with 2 gb.

My run times vary from around 13 minutes per WU (800 seconds) to 40 minutes per WU (2,400 seconds), depending on what other CPU jobs I am running, if they don't release the CPU for short periods then the GPU can't get enough CPU time to help process a WU and then slows down.
My CPU run time is under 100 seconds per WU on average (you can check this yourself as my computers are available to see).


I agree that the cpu time on mine nearly equal the run time. I was told when a second card is added it requires more cpu time, I have no idea, clearly that's not the case with yours. I believe it was mentioned earlier in this thread, I'm just going by what was mentioned.

The GPU can't process the whole work unit without some help from the CPU and other system resources.

The CPU, in order to run jobs, uses available main system memory as does the GPU.

I see that your computer has 2 GB of main memory.


Actually I have 4 1gb sticks installed.

I believe that this is your main bottleneck and why your jobs are taking so long to run.
With system resources limited the CPU is used more than normal due to the GPU making constant requests for those resources until it gets them.

An increase in main system memory would fix/improve your problem with reduced run times.

Both my computers have 4 GB of RAM (I know they are 32 bit so are using 3 GB), and also run all CPU Cores on various Boinc projects.
Some people give a complete core to a GPU, I don't do this but it would speed up my times if I did as the GPU would not have to wait for another project to release the CPU for the GPU to use it.
However I need all the CPUs so I sacrifice GPU speed for overall throughput.



How do I dedicate 1 core to the GPU? Currently I'm not crunching and cpu projects, only GPU.

Hope this information can help.

Conan


Your information is very much appreciated. If I can make this run faster, I'm all for it.

I did look at your WU's and I agree, you're processing must faster with less cpu usage, maybe because the cards aren't matched? One is 1gb and the other 2gb?



BOINC Start Up Information

[edit]



[edit/edit]

Moo on startup

10) Message boards : Number crunching : (2) 5870's (Message 1671)
Posted 10 Dec 2011 by Philadelphia
Post:
I 'think' I figured out what the problem with Moo was/is. Kidding on the 'I', even a blind squirrel can find an acorn.

I had booted/rebooted/booted/rebooted started boinc/stopped boinc/restarted/stopped it, etc, changed as many settings as carter has pills.

I had about 3 days worth of WU's to crunch and set the project to no new downloads.

I'm sure you could tell from some of my posts I was frustrated as hell.

In spite of my better judgement, I decided to 'allow new tasks', received 1 download, suddenly, instead of running two tasks it was running one task (running 0.05 cpu's + 2.00 ATI GPU's (device 0)) Before it was running 2 tasks with +1.00 gpu (device 0) and another (device 1)

It's still crunching the cpu but the run time is way down, I 'think', to something reasonable. If I can get more WU's I'll get a better idea.

My question (not to anyone in particular) is why did 'allowing new tasks' suddenly tell the computer to run one work unit instead of two and use both gpu's to do it, when stopping/starting BOINC didn't, even when it involved setting changes????

[edit] I ran about 20 WU's since and everything is fine, what a strange situation. Maybe the project guru has input?
11) Message boards : Number crunching : (2) 5870's (Message 1661)
Posted 9 Dec 2011 by Philadelphia
Post:
Going to clear out my Moo's cashe and move on, sucks, I was expecting about 700k a day :(


Dan it was only 7 hours between your first post and your last post, give us a chance to take some naps and analyze the data guy.

Now you know that Moo is a project that uses BOTH of your gpu's on ONE workunit, right?! This could be what you are seeing the 'stream'ing of the data between the cards so ALL of BOTH cards resources are being used on ONE workunit. When you only have one card no 'stream'ing is needed, so the numbers are zero. Remember too your two cards are NOT identical, one has 1gig of memory while the other card has 2gig of memory, this could be why the workunits are taking so long. You MIGHT be able to limit the resources thru config settings but don't quote me on that.


That is true, they are not identical

I was not aware that Woo used both cards to crunch 1 WU since two of them are running at the same time, or at least in BOINC they show two running at the same time.


12) Message boards : Number crunching : (2) 5870's (Message 1657)
Posted 9 Dec 2011 by Philadelphia
Post:
Going to clear out my Moo's cashe and move on, sucks, I was expecting about 700k a day :(
13) Message boards : Number crunching : (2) 5870's (Message 1656)
Posted 9 Dec 2011 by Philadelphia
Post:
Having a great conversation with myself here, lol, kidding. :) Just put the other one back in and woooolaaaaa $^#&#

dnetc files going crazy again.

Guess I'll just keep one out until it can be figured out :( wasted 300K a day :(

14) Message boards : Number crunching : (2) 5870's (Message 1655)
Posted 9 Dec 2011 by Philadelphia
Post:
I just pulled the second 5870 and the result is back to what it was before, I'm pulling my hair out :(

15) Message boards : Number crunching : (2) 5870's (Message 1654)
Posted 9 Dec 2011 by Philadelphia
Post:
I just pulled the second 5870 and that dnetc518 file is back to 0 cpu.

Thoughts?

16) Message boards : Number crunching : (2) 5870's (Message 1652)
Posted 8 Dec 2011 by Philadelphia
Post:
I was running one 5870 with WU's in the 1,700 run time and 14 in cpu time.
Now I installed another one and the WU's are 2000 to 3000 run time and just short of that for cpu time too. I also noticed that two dnetc518 are running about 50% cpu usage, I don't remember the dnetc518 file running that high before although I can't swear to it. I may just pull one 5870 out and see what happens.

Here are several screen captures, sorry for the size but I wanted to share the information, it may be helpful.

BOINC SETTINGS



MOO SETTINGS



BOINC ON STARTUP SHOWING BOTH GPU'S



WU'S RUNNING SHOWING BOTH GPU'S



DNETC518 FILE IN TASK MGR



BEFORE AND AFTER ADDING THE SECOND GPU, SMALL NUMBER ARE WITH 1, LARGE WITH BOTH



Thoughts/suggestions?

Dan/Philly
17) Message boards : Number crunching : dnetc518-win32-x86-stream.exe (Message 1611)
Posted 3 Dec 2011 by Philadelphia
Post:


That one has nothing todo with Philadelphias problem nor with the dnetc client(which was coded using ATI CAL/STREAM NOT openCL).

That one -> dnetc518-win32-x86-stream.exe is the CPU app and of course it'll utilize the CPU at 100%.

@Philadelphia, just make sure to disable receiving CPU tasks in your account settings and that will solve the issue.


I changed the settings to these settings. Ok, or modify?

Thanks.

Hah! We see only our own settings, not yours, with that URL. Just follow Cunch3er's instructions and you'll be fine. ;)

If you want to see results right away, delete any tasks with that particular app.


Gotcha. The second 5870 is out at the moment to make a dummy plug. I drove right by a Radio Shack and forgot to stop. :(

Thanks for everyone's help!

Dan/Philly
18) Message boards : Cafe : Last person to post win #2 (Message 1600)
Posted 3 Dec 2011 by Philadelphia
Post:
That just gave me a headache :)
19) Message boards : Number crunching : dnetc518-win32-x86-stream.exe (Message 1596)
Posted 2 Dec 2011 by Philadelphia
Post:


That one has nothing todo with Philadelphias problem nor with the dnetc client(which was coded using ATI CAL/STREAM NOT openCL).

That one -> dnetc518-win32-x86-stream.exe is the CPU app and of course it'll utilize the CPU at 100%.

@Philadelphia, just make sure to disable receiving CPU tasks in your account settings and that will solve the issue.


I changed the settings to these settings. Ok, or modify?

Thanks.
20) Message boards : Number crunching : dnetc518-win32-x86-stream.exe (Message 1591)
Posted 2 Dec 2011 by Philadelphia
Post:
There is a long standing bug in OpenCL applications that affects dual GPUs (either on one card or two single GPU cards). The bug grabs a complete CPU Core for each GPU running. There is nothing the Project can do about it - has to be fixed at the OpenCL/AMD/NVidia side of life.

Its slowly getting resolved, its been around for a few months now - only affects those with more than one GPU running on a box. AMD recently started testing a Linux fix, its not yet known when the Windows fix will be out. Spring next year would be a crude guess - its a nasty one to resolve apparently.

Regards
Zy


Thanks for sharing the information, I'd really not like to have my 4 year old processor crunching 100%, it already has 1 1/2 years of 100% crunching when I first got it.

I guess I'm just going to run one of them for Moo than, and one for another project, unless it's an issue regardless if the cards are split between projects.



Next 20


 
Copyright © 2011-2024 Moo! Wrapper Project