Welcome to MilkyWay@home

GPU app teaser

Message boards : Application Code Discussion : GPU app teaser
Message board moderation

To post messages, you must log in.

Previous · 1 . . . 8 · 9 · 10 · 11 · 12 · 13 · 14 . . . 19 · Next

AuthorMessage
Profile Kevint
Avatar

Send message
Joined: 22 Nov 07
Posts: 285
Credit: 1,076,786,368
RAC: 0
Message 11896 - Posted: 20 Feb 2009, 21:36:46 UTC - in response to Message 11885.  

Question...

And sorry if this has been brought up before -

The other BOINC projects do not run alongside the GPU app.

Is this correct, or is there something I am missing ???

Do you get errors if you try to fully load the GPU and the remaining cores with different projects or does the BOINC client simply refuse to run another project at the same time as MW?
I thought I have seen some guys are running several projects along with the MW GPU app.

Edit:
Beremat provided a screenshot in this thread showing he is at least attached to several projects at once. Maybe he can answer if he was able to run them at the same time.
I got a report that one gets a VPU recover error if other projects run with the MW GPU app creating a 100% CPU load. He said a possible workaround is to configure BOINC to use only 3 cores on a Quadcore. Obviously noone tried to simply deactivate the VPU recover feature yet. May be worth a shot.


Nope no errors at all - only MW WU's crunch, until the cache of MW is empty, then the other project kicks in.

I am currently running 8 MW WU's at a time -

I only am seeing about 25-30% CPU load.



.
ID: 11896 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Temujin

Send message
Joined: 12 Oct 07
Posts: 77
Credit: 404,471,187
RAC: 0
Message 11898 - Posted: 20 Feb 2009, 21:48:16 UTC - in response to Message 11817.  

2. I had to install Microsoft .NET Framework v2.0 for x64. It didn't like that I had .NET v3.5 installed which I thought was supposed to be an extension to 2.0. It wanted 2.0 as well.

I didn't read this line properly.
I already had .net framework 2 installed and thought that was all I needed.
After numerous re-installs of the catalyst drivers & boinc I realised I should also install .net framework 3.5
After it rebooted all is working :)

WOOHOO

running 4 WUs at a time with each taking 13 seconds

Stunning work Cluster Physik


just to recap
I needed both .NET Framework V2.0 and V3.5
ID: 11898 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Slicker [TopGun]

Send message
Joined: 20 Mar 08
Posts: 46
Credit: 69,382,802
RAC: 0
Message 11902 - Posted: 20 Feb 2009, 22:42:39 UTC - in response to Message 11898.  

2. I had to install Microsoft .NET Framework v2.0 for x64. It didn't like that I had .NET v3.5 installed which I thought was supposed to be an extension to 2.0. It wanted 2.0 as well.

I didn't read this line properly.
I already had .net framework 2 installed and thought that was all I needed.
After numerous re-installs of the catalyst drivers & boinc I realised I should also install .net framework 3.5
After it rebooted all is working :)

WOOHOO

running 4 WUs at a time with each taking 13 seconds

Stunning work Cluster Physik


just to recap
I needed both .NET Framework V2.0 and V3.5


Just installed a second card on an XP64 box w/ a Q9450. I ONLY installed .NET 2.0 SP1 and it worked fine. Maybe the issue is that the catalyst driver needs SP1? .NET 3.5 is NOT installed on the box. Hmmm....

Glad it worked for you though.
ID: 11902 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Slicker [TopGun]

Send message
Joined: 20 Mar 08
Posts: 46
Credit: 69,382,802
RAC: 0
Message 11903 - Posted: 20 Feb 2009, 22:45:56 UTC

New install on a Q9450 running XP64 w/ Boinc 6.4.5. It runs 8 gpu WUs at a time. Nice. But, it will only download 12 WUs as if it only recognizes that it has 1 cpu rather than 4 cores. How can I get it to download 12 per core? Or, is it because it is only running on 1 cpu (no other projects) that it will only download 12 at a time?
ID: 11903 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile GalaxyIce
Avatar

Send message
Joined: 6 Apr 08
Posts: 2018
Credit: 100,142,856
RAC: 0
Message 11906 - Posted: 20 Feb 2009, 22:56:52 UTC - in response to Message 11794.  
Last modified: 20 Feb 2009, 22:57:31 UTC

what is the latest version of the gpu app? I'm currently using 0.17 and not sure if their is an update i should be using. If there is any reason to upgrade or known problems?

The latest version would be 0.19 (the link is some posts above), but nothing changed besides a CPU detection like in the CPU version 0.19. A GPU detection would have made more sense ;) So I would regard an upgrade as optional and not needed atm.
Well, 32Bit Win systems are supported now, too.

Actually the 32Bit app still needs some testing, the known problems are quite similar to the 64Bit version I guess. But according to some reports, it may not be such a problem to run GPU-Z when the GPU is used as it is with the 64Bit version. But that's more of a speculation.

If the ones running the app provide some feedback about their settings and the stability, it would be great.

Windows 32/64 ATI version 0.19 are now in zslip.com and zslip.net

Thanks again Cluster Physik

ID: 11906 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Paul D. Buck

Send message
Joined: 12 Apr 08
Posts: 621
Credit: 161,934,067
RAC: 0
Message 11913 - Posted: 20 Feb 2009, 23:32:36 UTC

Well, Frys had a 4870 on sale for $199 with discount and rebate... kinda hard to turn that down ... so ...

Installed the card, 8.12 drivers ... not sure if I had .Net versions installed (am installing 32 patches as I type this, one of which is the 3.5 version if I read correctly) ... Using BOINC version 6.5.0 though that does not seem to affect things significantly (one way or the other).

it seems to be running successfully with this caveat ... if I allow any other project I get fewer and fewer tasks running on the system.

I tried using configuration settings to restrict the number of cores in use to 3 (of 4 virtual CPUs) but that does not seem to help, what that seems to do is to cut the number of Milky way tasks to 6 vice 8 ...

I have been up and down this thread several times and I think I am still missing something ... am I supposed to be adding another configuration file other than the app_info?

SOMEWHERE I thought I read a note about changing one of the configuration files to allow parallel execution of the GPU tasks with CPU tasks from other projects ...

So, I guess what I am asking, has anyone been able to run MW GPU while running other projects?
ID: 11913 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
Message 11920 - Posted: 20 Feb 2009, 23:46:46 UTC - in response to Message 11913.  
Last modified: 20 Feb 2009, 23:49:15 UTC

So, I guess what I am asking, has anyone been able to run MW GPU while running other projects?

Beremat has answered this. He has a HD4850 paired with a C2D 8400 running WinXP32.

I was running a few projects before, and I was not getting any errors whatsoever. It seemed as though 2 other projects' WUs could run alongside. One would get 100% of one core, and the other would only get some percentage of the other core when MW wasn't using the CPU.
ID: 11920 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Paul D. Buck

Send message
Joined: 12 Apr 08
Posts: 621
Credit: 161,934,067
RAC: 0
Message 11938 - Posted: 21 Feb 2009, 0:29:55 UTC
Last modified: 21 Feb 2009, 0:44:12 UTC

I wonder what he did ... cause it does not work for me ... if I start up any other project, even FreeHal I lose one task on the ATI GPU ... so, for example I have FreeHAL enabled and only have 7 tasks running on the GPU ... if I suspend that task I get 8 on the GPU ... if I allow normal operation I get no GPU activity...

If I allow one CPU task to run I lose two GPU tasks ... system loading is 65% ... running one FreeHal task and two VP tasks load is 70%-80% but only 4 GPU tasks ...

Very strange ...

I wonder if he made a, what is it, cc_config file?

{edit}
Hmm, that did not work as expected. If I make a CC_Config I can set the CPUs up to higher numbers, and if I set it to 5 I can run one task in parallel with 8 MW GPU tasks ... but if I set it to 8, I can have 4 tasks running as normal ... but, I have to manually manage the tasks, turning them on one at a time to prevent more tasks running than expected ...
ID: 11938 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
Message 11949 - Posted: 21 Feb 2009, 1:28:48 UTC - in response to Message 11938.  

I wonder what he did ... cause it does not work for me ... if I start up any other project, even FreeHal I lose one task on the ATI GPU ... so, for example I have FreeHAL enabled and only have 7 tasks running on the GPU ... if I suspend that task I get 8 on the GPU ... if I allow normal operation I get no GPU activity...

If I allow one CPU task to run I lose two GPU tasks ... system loading is 65% ... running one FreeHal task and two VP tasks load is 70%-80% but only 4 GPU tasks ...

Very strange ...

Not really. Running fewer tasks concurrently on the GPU does not lower the througput, it is perfectly okay.

And remember that BOINC is not aware that the MW GPU app actually uses the GPU (no ATI support yet). It is scheduled like a normal CPU application. On newer versions (6.2 and up I think) it obeys the avg_ncpus and max_ncpus values in the app_info.xml. They are set to 0.5 and 1 respectively, to allow for an easier overlap of at least two WUs (increases efficiency and GPU load slightly, but more than two do not help) also with a lower resource share. But one can also play around with the values. But this actually just influences how many tasks BOINC schedules to run, not how much of a core is really used. You can save some CPU time on the GPU application (effectively freeing up a bit of the one full core the GPU app takes irrespective of the number of concurrently running MW WUs) if you lower the values and reduce the MW resource share accordingly.

Just an example, you have a quad core, the MW resource share is at 25% and some other project is at 75%. BOINC will then run two MW WUs concurrently (using one core) and three additional WUs from that other project with the default settings in the app_info.xml. That's perfectly okay, as the throughput of the GPU app does not depend on the number of concurrent WUs (as longs as two or more are running, but even that is only a 10% effect, even less on the slower HD3800 series). After all, you have only one GPU. It's almost like running multithreaded on a single core. That scenario will give you 100% CPU load and still the maximal throughput of the MW GPU application.

Now reduce the avg_ncpus values in the app_info to 0.25 (the max_ncpus should stay at 1). With the same resource share settings as above (25/75) there will still be 3 WUs from the other project, but now 4 MW units. That should still give the same troughput as with the 2 WUs from above. But now you could also reduce the MW share a bit to let's say 13/87. The BOINC scheduler will run in most cases 4 WUs of the other project and only 2 MW WUs. Effectively 5 active tasks (as I explained basically all MW tasks together counts as only one in this respect no matter how many WUs actually run) battle for the 4 cores of the system. The throughput of the other project should rise slightly with his approach and MW should not take too much of a hit (has to be tested though).

Hope this may shine some light on the issue.
ID: 11949 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Brickhead
Avatar

Send message
Joined: 20 Mar 08
Posts: 108
Credit: 2,607,924,860
RAC: 0
Message 11961 - Posted: 21 Feb 2009, 2:03:40 UTC

Should I change the avg_ncpus from the default value when using a dual-GPU card (4870x2)?
ID: 11961 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Paul D. Buck

Send message
Joined: 12 Apr 08
Posts: 621
Credit: 161,934,067
RAC: 0
Message 11964 - Posted: 21 Feb 2009, 2:08:28 UTC

Some light, though I think I am drowning ... still ...

Ok, it looks like I am going to have to fiddle with the files ...

Though I don't have a lot of experience with the operation (2 hours or so) I think I will have to play a bit ...

I think till tomorrow when I have the courage to attack this I will just let it run and if it runs dry (as it has), it runs dry for a bit. Though based on skimpy observation, running fewer tasks on the GPU does not appear to improve the throughput ...

In other words, the GPU is not saturated at 8 tasks ... however, I may be saturating the supporting CPU with more tasks running ... Ok, I think I see what is going on here ...

For the moment, tonight I am going to run with my CPDN task and that allows 6 GPU tasks to run and tomorrow I will look at other settings ...

Thanks for the explanation, not sure I understand it ... but ... that means it is probably me not getting something simple ...
ID: 11964 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
Message 11965 - Posted: 21 Feb 2009, 2:14:19 UTC - in response to Message 11961.  

Should I change the avg_ncpus from the default value when using a dual-GPU card (4870x2)?

In the moment multiple GPUs are not supported (but hopefully I will find some time to put it in). In principle the app should use just one of the GPUs on your card. With multiple cards in a system it is the same.
But could you please test that it runs at all?
Editing of the app_info.xml is not needed if all is running well.

By the way, a single HD4870 is already that fast. Do you imagine what two of them would do? :o
ID: 11965 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
Message 11966 - Posted: 21 Feb 2009, 2:24:46 UTC - in response to Message 11964.  
Last modified: 21 Feb 2009, 2:31:03 UTC

Some light, though I think I am drowning ... still ...

Ok, it looks like I am going to have to fiddle with the files ...

Though I don't have a lot of experience with the operation (2 hours or so) I think I will have to play a bit ...

I think till tomorrow when I have the courage to attack this I will just let it run and if it runs dry (as it has), it runs dry for a bit. Though based on skimpy observation, running fewer tasks on the GPU does not appear to improve the throughput ...

In other words, the GPU is not saturated at 8 tasks ... however, I may be saturating the supporting CPU with more tasks running ... Ok, I think I see what is going on here ...

For the moment, tonight I am going to run with my CPDN task and that allows 6 GPU tasks to run and tomorrow I will look at other settings ...

Thanks for the explanation, not sure I understand it ... but ... that means it is probably me not getting something simple ...

So let's make it simple.

A HD4870 finishes about 400 WUs (of the short type) per hour. It does so if you run 2 WUs at a time, and it still arrive at the same throughput as if you would run 4, 8 or even 16 WUs concurrently on that single card. The throughput does not rise like on a a multi core CPU (there is only one GPU).

So just fire up some other projects. As longs as there is at least one running MW task left, you will generate one result every 9 seconds (wall clock time).

And with less concurrent MW tasks you get an even better impression of the speed ;)
ID: 11966 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
bobgoblin

Send message
Joined: 8 Dec 07
Posts: 60
Credit: 67,028,931
RAC: 0
Message 11979 - Posted: 21 Feb 2009, 3:01:01 UTC - in response to Message 11793.  

I upped the limit to 5000, its probably not enough :P Didn't want to jump all the way to 10k and have craziness start happening.

You don't need to raise it until the multiple GPU support is working. 10,000 WUs a day on a dual core are just enough for a HD4870 at stock speed.


and with the longer wu's 5k is probably enough.

though I noticed on the i7:

Maximum daily WU quota per CPU 4999/day

and it was only 4998/day when I looked earlier while all the other machines showed 5000.

and since we've gotten longer work units, i've been having problems with the GPU app freezing after about 4 hours - i7 with ati 4800. so, this morning, I updated to CCC 9.1. Then noticed it froze after 2 1/2 - 3 hours when i checked it from my phone.

when I got home tonight, it didn't look like 9.1 installed correctly. So I completely uninstalled and reinstalled 9.1, then it would freeze up after an hour or so. I just installed the new .19 gpu app, so maybe that will clear it up?

but i wonder if b/c these are longer wu's, that I'm just putting too much strain on the gpu continuously crunching 16 wu @ a time?
ID: 11979 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Brickhead
Avatar

Send message
Joined: 20 Mar 08
Posts: 108
Credit: 2,607,924,860
RAC: 0
Message 11982 - Posted: 21 Feb 2009, 3:04:43 UTC - in response to Message 11965.  

But could you please test that it runs at all?

Oh yes, the app works just as intended, I guess, chewing through one WU every 8-12 seconds (depending on the WU).

By the way, a single HD4870 is already that fast. Do you imagine what two of them would do? :o

Having two GPUs available, and knowing that one of them will remain idle for now, imagining is just about all I do. But hey, at some 50 times the speed of one 4.0 GHz Yorkfield core (stock app), I'm already laughing :D
ID: 11982 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
[B^S] Beremat

Send message
Joined: 19 Feb 09
Posts: 33
Credit: 1,134,826
RAC: 0
Message 12015 - Posted: 21 Feb 2009, 3:56:06 UTC - in response to Message 11979.  
Last modified: 21 Feb 2009, 4:18:58 UTC

[quote]I upped the limit to 5000, its probably not enough :P Didn't want to jump all the way to 10k and have craziness start happening.

and since we've gotten longer work units, i've been having problems with the GPU app freezing after about 4 hours - i7 with ati 4800. so, this morning, I updated to CCC 9.1. Then noticed it froze after 2 1/2 - 3 hours when i checked it from my phone.

when I got home tonight, it didn't look like 9.1 installed correctly. So I completely uninstalled and reinstalled 9.1, then it would freeze up after an hour or so. I just installed the new .19 gpu app, so maybe that will clear it up?

but i wonder if b/c these are longer wu's, that I'm just putting too much strain on the gpu continuously crunching 16 wu @ a time?

Hm, I doubt that you're overstressing your card. It is likely that there is a fault with the drivers somewhere. I've been crunching for around 46 hours straight on my 4850 with Catalyst 8.9, and no errors apart from when I was gaming.
I suggest you rollback to 8.9 and see how it goes.

EDIT: Oh! Another thing. Monitor your temperatures on your GPU while crunching if you can. I had to increase the fan speed on my 4850 to 50% because I was afraid that 85C was a bit much. Even this slight increase from 30 to 50% cooled the card all the way down to 63C stable.


Also, note that I haven't done a single WU on MilkyWay with my CPU ever. All the results you see on the left are with one single GPU in 46 hours, just under 2 days. Subtract around 8 hours because I was gaming. That makes around 38 hours.
100,000 credits\38 hours=2.6k credits/hr on a 4850, imagine what a 4870 can do!

ID: 12015 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Paul D. Buck

Send message
Joined: 12 Apr 08
Posts: 621
Credit: 161,934,067
RAC: 0
Message 12037 - Posted: 21 Feb 2009, 5:53:01 UTC - in response to Message 11966.  

So let's make it simple.

A HD4870 finishes about 400 WUs (of the short type) per hour. It does so if you run 2 WUs at a time, and it still arrive at the same throughput as if you would run 4, 8 or even 16 WUs concurrently on that single card. The throughput does not rise like on a a multi core CPU (there is only one GPU).

So just fire up some other projects. As longs as there is at least one running MW task left, you will generate one result every 9 seconds (wall clock time).

And with less concurrent MW tasks you get an even better impression of the speed ;)


I guess I was seduced by the image of 8 tasks running at the same time ... With two GTX 295s in one of my other systems it is easy to get seduced by the thrill of speed ...

I guess that I was also getting suckered by the fact that I did not have the debts right so the MW tasks were not running. But, now the system seems to be content and is running the tasks through quite nicely ... though I only have one in flight at a time ... so I will see how it goes ...

Thanks for the help ... what the heck ... it seems to be running ... now I have to look at my other systems to make sure they are still cooking along too ...

Thanks again ...
ID: 12037 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Exar Kun [HoloNet]
Avatar

Send message
Joined: 12 Nov 08
Posts: 26
Credit: 1,542,686
RAC: 0
Message 12043 - Posted: 21 Feb 2009, 8:09:20 UTC

Upgraded to GPU app version 0.19.

Works fine on Vista 64, 4 Go RAM, HD 4850 512 Mo, Core2Duo.

It's running 4 MW units (0.50 CPU each), and 1 World Community Grid unit.

The two core are not working at 100%. In version 0.17 I was running two WCG units, and two MW units, and the 2 cores was at 100%...

Little problem here : when I attached a new WCG project, the 4 MW stopped (the calculation seems to work, but the progress bar is freezed for the 4 units).

A BOINC shutdown almost solved the problem : I have now 1 WCG unit and 2 MW units. Haaaa it changed again : 2 WCG units, and only 1 MW unit !!!

:)
Star Wars BOINC Team



ID: 12043 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
bobgoblin

Send message
Joined: 8 Dec 07
Posts: 60
Credit: 67,028,931
RAC: 0
Message 12065 - Posted: 21 Feb 2009, 12:40:37 UTC - in response to Message 12015.  
Last modified: 21 Feb 2009, 12:43:47 UTC


I suggest you rollback to 8.9 and see how it goes.

EDIT: Oh! Another thing. Monitor your temperatures on your GPU while crunching if you can. I had to increase the fan speed on my 4850 to 50% because I was afraid that 85C was a bit much. Even this slight increase from 30 to 50% cooled the card all the way down to 63C stable.


i left it at 9.1 but took your advice of upping the fan speed. that's brought it down from ~83c to ~77c.

and it made it through the night without locking up. So it could be that it was overheating. But I had also upgraded to version .19. So, was there a change in there that corrected the problem?

either way, v.19 is working fine on an i7 and hd4870 with 512m
ID: 12065 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
Message 12067 - Posted: 21 Feb 2009, 13:12:04 UTC - in response to Message 12065.  

i left it at 9.1 but took your advice of upping the fan speed. that's brought it down from ~83c to ~77c.

and it made it through the night without locking up. So it could be that it was overheating. But I had also upgraded to version .19. So, was there a change in there that corrected the problem?

either way, v.19 is working fine on an i7 and hd4870 with 512m

Besides the CPU detection nothing changed between 0.17 and 0.19. So maybe really a temperature problem.

But I have seen your crunch times are slightly on the high side. This could be caused by running too many WUs concurrently on the GPU. At a certain point the RAM on the graphics card is not sufficient for the number of WUs taking space there. Before it errors out (when even more WUs would be crunched), it slows down (probably some swapping over PCI-Express happens). And with 16 WUs it is getting already a bit crowded on a 512MB card.
Another reason for the higher times could be that the card runs downclocked in a power saving mode. Maybe you should check the clock speed of the card.

Furthermore you may think about attaching to a second BOINC project with that i7. This will reduce the number of MW WUs that are running a he same time, but not the throughput. You will still finish the same number of WUs per hour even with less concurrently running WUs. In fact, it could even rise in your case. Furthermore your CPU cores wouldn't be idling that much ;)
ID: 12067 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Previous · 1 . . . 8 · 9 · 10 · 11 · 12 · 13 · 14 . . . 19 · Next

Message boards : Application Code Discussion : GPU app teaser

©2024 Astroinformatics Group