GPU task ?

S
Simon ID: 1264 Posts: 45
31 Mar 2013 02:19 PM

I'm running Boinc with CE as the project manager. I have had problems with my diplays keep flickering and a message coming up saying the graphics driver had reset. Looking in boic i noticed a CE task that is running on the GPU with 0.6something CPU. I thought Ce did not do GPU tasks ?

Has there been a change, are we experimenting ? I also had C++ runtimes crash or something.

J
Jo1252 ID: 1311 Posts: 81
01 Apr 2013 01:04 AM

I am in the same situation, and my firewall keeps annoying me with the gpu process to access the internet. Although I fixed that by allowing it to access the internet at all time.

computer specs:

nvidia card (GTX 560 Ti) and latest driver (314.22).

Computer on windows 7 x64 with 8gb ram.

AMD phenom II X6.

Hope you can fix this issue soon.

p.s: I just noticed my card is slightly overclocked by the manufacturer, it might be the source of the problem. I'll try later to underclock it to standard by nvidia.

S
Simon ID: 1264 Posts: 45
01 Apr 2013 09:16 AM

Yes I had the firewall keep asking permissions too

Matt ID: 44 Posts: 256
01 Apr 2013 03:35 PM

simon, just curious -- do you know what kind of GPU you have?

S
Simon ID: 1264 Posts: 45
01 Apr 2013 03:37 PM

Geforce GT520

S
Simon ID: 1264 Posts: 45
01 Apr 2013 03:38 PM

Desktop version

S
Simon ID: 1264 Posts: 45
01 Apr 2013 03:41 PM

I'm also on latest drivers

Matt ID: 44 Posts: 256
01 Apr 2013 03:41 PM

btw -- yes, we have a GPU app running on machines with some of our commercial partners (*where environment is more standardized).  we thought to allow this on public machines, where possible -- ie, users who have a generic BOINC client.

but likely we'll suspend this application soon as we're finding it -- so far -- a bit hard to optimize for the wider spectrum of device types out "in the wild".  (*thought we'd made some progress here but maybe not.)

Matt ID: 44 Posts: 256
01 Apr 2013 03:42 PM

meanwhile -- anyone having troubles, pls just "suspend" this project in your desktop applications.  you'll keep earning prize draw credits in the other projects.

S
Simon ID: 1264 Posts: 45
01 Apr 2013 03:43 PM

Can't you target specific GPU's ? and gradually work from there. I don't know if for example the nvidia GPU has a standard cuda interface that you could work with allowing for the less capable processors, there are various levels of "cuda" depending on the GPU

Matt ID: 44 Posts: 256
01 Apr 2013 03:52 PM

yes, we have been targeting GPUs.  expanded the range of devices this weekend -- which is where you started getting jobs, it seems.  apparently more tuning to go...  sorry for any inconvenience...

S
Simon ID: 1264 Posts: 45
01 Apr 2013 03:54 PM

No problem, it just blanks screens for a bit flickers and recovers, but you loose all task work and it just keeps crashing so they seem to have a problem at the same points.

 

what are you using as a target ? is this a problem where you ideally need a higher level cuda GPU ?

Garrett B ID: 3819 Posts: 8
01 Apr 2013 04:00 PM

I personally love the gpu tasks, I've been running the generic Boinc client for this reason. I run Einstein@Home (Supended the rest) I am able to complete a 140,000 G/Flop task on my GTX 480 every 20-25 minutes. I love the "feature" to run gpu able tasks. I say feature with quotes since this is provided via Boinc client. 

Please keep these gpu task avalible otherwise my compulation falls near 0 with my cpu, at least in comparison to my graphics card.

-Thanks

Matt ID: 44 Posts: 256
01 Apr 2013 04:04 PM

our initial GPU application is particularly demanding.  there are a range of interrelated factors that affect targeting, including: GPU, onboard memory, and driver versions.  Also, more than other kinds of applications, GPU apps seem especially sensitive to details of what the user is doing on his/her machine...  so we're working on it...

Long term, we're commited to GPU computing, as it seems to be what commercial clients want (*and that's what makes CE go); so we'll soon be releasing a new version of the CE desktop app to enable this for all users by default.

S
Simon ID: 1264 Posts: 45
01 Apr 2013 04:09 PM

Yes GPU computation rocks, although I find my computer goes sluggish so only usually let it run when idle, but the CE task was not causing me problems so you made something good there

S
Simon ID: 1264 Posts: 45
10 Apr 2013 07:30 AM

I have a Quadro 4000 card at work ewith 2GB of Ram. Can you tell us more about the GPU's you are targetting and any driver settings ? The driver for my quadro has various profiles for example.

I'm also getting C++ copout errors that affected CPU tasks as well. Any ideas ?

S
Simon ID: 1264 Posts: 45
10 Apr 2013 08:00 AM

I have just successfully completed a CE GPU task on my Quadro 4000 after installing latest drivers with the performance option set in profiles if that helps.

J
Jo1252 ID: 1311 Posts: 81
10 Apr 2013 11:02 AM

I bet it was fortran errors.

climate prediction had a bad bunch of work units recently.

it might have been this, if you have more information we could help better.

I'm glad that you can use your gpu!

S
Simon ID: 1264 Posts: 45
10 Apr 2013 01:02 PM

I've got problems again it runs but as the task completes the video driver crashes, screens flicker and the task is shown as completed and starts uploading. I don't know if you can check to see if my tasks have complted successfully as this seems the case to me although it seems I have to pay for that with a driver crash every 7 minutes.

This is on my Quadro 4000 with 2GB of VRAM

J
Jo1252 ID: 1311 Posts: 81
10 Apr 2013 03:55 PM

I also have the driver crash, but the work seems to be ok.

it always happen at the end of the work, as if there was a problem exiting the app.

if you find it too annoying, you can suspend the project.

S
Simon ID: 1264 Posts: 45
10 Apr 2013 05:30 PM

Yea I opted for the screen flicker every 7 minutes, hopefully someone can tell us that the effort is worth it

J
Jo1252 ID: 1311 Posts: 81
10 Apr 2013 05:32 PM

Take a look at your stats, if they rise it's because the work is validated.

S
Simon ID: 1264 Posts: 45
10 Apr 2013 06:06 PM

I guess so, how long does it take ? I left my work compyter on over night. It should have racked up a fair amount by the morning. It churns 140000 whatever flop jobs for eistein at home in 30 minutes or less

J
Jo1252 ID: 1311 Posts: 81
10 Apr 2013 06:09 PM

What I meant is look at the charity engine stats in the project tab.

Look at the work done, and if the number goes up, it means the work is done and validated (no error).

Paul Mackinnon ID: 3677 Posts: 3
12 Apr 2013 04:49 AM

I'm also experiencing the combination of excessive display driver crashes and resets on W7 x64, as well as the incessant firewall requests from CE related exe's which appear to be linked to bitcoin mining operations, either that or they just share related process names and such.

Using a GeForce 560Ti here, with the latest drivers.

Matt ID: 44 Posts: 256
14 Apr 2013 09:29 PM

At some point this week we'll be stop sending the current CE app to a number of GPU types which seem to having problems-- until we can better tailor things to different hardware setups.  Sorry to those of you who have been inconvenienced.

G
Graham ID: 2422 Posts: 27
15 Apr 2013 06:59 AM

Aha, this thread may explain why my daily average has shot up and almost all the work seems to be on Einstein now.  However all this talk of NVidia cards makes me unsure.

I have an AMD 6950 running in my machine.  Do you have tasks that run on AMD cards?

 

S
Simon ID: 1264 Posts: 45
15 Apr 2013 07:04 AM

probably not. ATI/AMD and nVidia have two seperate environments that run these sorts of applications so you need to code for both. As stated this works only on some nVidia cards so I doubt that anything has been written for ATI cards yet. Most power users tend to get nVidia i think. If you look at the main boinc website you will see that more projects support nVidia than ATI and they are specifically written for one or the other. It is also possible that the two GPU types have different weeknesses and strengths so each may be better from certain projects

Rytis Slatkevičius ID: 195 Posts: 56
15 Apr 2013 07:04 AM

We do have a preliminary application for ATI/AMD cards too, however, we are now reorganizing what we send to public as we do not want the testing applications to negatively affect you, our public members.

S
Simon ID: 1264 Posts: 45
15 Apr 2013 07:06 AM

Your best bet is to gradually work up your skill set for each proccessor and make a way of having tasks sent only to those GPU's that way you can work with compatible GPU's and not affect the others and expand gradually.

G
Graham ID: 2422 Posts: 27
15 Apr 2013 10:44 AM

Thanks for that Simon.  Yes I believe the Nvidia cards are better at floating point than AMD but the latter are better at things like hashtable calculations.  I will await developments.:)

S
Simon ID: 1264 Posts: 45
15 Apr 2013 11:08 AM

I'm dreaming up my new computer and am torn between the fastest 4 core AMd I can get a slighly slower and more costly AMD with video processor included for things like boinc as I'd still get an nVdia card and that way have both core types available. Things may not work the way I think of course.

I've noticed that few programs run on more than one core (I use solid edge at work that costs thousands and we run it on an 8 core machine, so we only get to use 12.5% of it's power when we need speed the most) so my preference is for less and faster.

 

J
Jo1252 ID: 1311 Posts: 81
16 Apr 2013 01:20 AM

I noticed I do not receive any new GPU task from charity engine so I bet you do not send anymore to troublesome GPUs.

If I can do a suggestion, it would be awesome if you could make a page where infos on the charityengine project are (exemple: supported GPUs at the moment, what the app does (data mining, simulating this,calculating that,etc.), and any other useful information. I do understand that some work our computer do is confidential, in such a case just tell broadly what it does.)

Thank you for reading me.

Mark McA ID: 179 Posts: 213
16 Apr 2013 03:12 AM

Hi Jo,

Problem with listing everything is that some people will always want to support some tasks more than others, so they will alter the CPU share or even delete every project except their favourite.

That messes everything up, because we can't offer a computing service unless every PC behaves in (more or less) the same way. It would be too unreliable.

We read every comment, by the way. Thank you for writing!

Cheers,

Mark

Matt ID: 44 Posts: 256
16 Apr 2013 03:21 AM

Btw, as for app specs -- we're currently in a transition stage, but within a few months we'll have a range of applications online, and most machines will be able to run at least one.  Job provisioning will be handled automatically.