Hello,

I am using BOINC GUI 6.4.4 on my Debian machine. It is equipped with a Geforce 8400 GS which works under CUDA, for example on Windows XP Professionnal.

However, when launching BOINC under Debian, it displays ``CUDA devices found'', then ``Coprocessor: GeForce 8400 GS (1)'' but SETI's apps, or GPUgrid's apps never download CUDA tasks, and I am thus unable to compute for CUDA tasks on my Debian machine.

What can I do? Thanks.

---

See http://www.student.montefiore.ulg.ac.be/~merciadri/ for more pieces of info.

what driver are you running?, some of these projects have a minimum driver requirement. How much memory does the card have?

As far as I know, SETI doesn't have a linux application. There are only 3rd-party optimised apps available.

Gruß,
Gundolf

---

Computer sind nicht alles im Leben. (Kleiner Scherz)

I am using Debian's default driver.

The card has 256 MB ram but works for both SETI, F@h, etc., under XP Pro.

Gundolf, do you want to say that SETI does not have any SETI GPU application?

Thanks.

what is the nvidia driver version number?

178? 180.XX????

this may be your problem.

in seti prefernces on your account page, you need to have cuda, use gpu checked.

fyi, this is the wrong area to post this in, you need to post in questions, and problems area. it's ok, ageless will move it one day when he sees it.

Gundolf, do you want to say that SETI does not have any SETI GPU application?

Not for linux, yes. At least according to their Applications page.

Gruß,
Gundolf

gpugrid does however have a linux app. but you need to have the required drivers.

gpugrid does however have a linux app. but you need to have the required drivers.

But I read on various threads (on the Internet) that the message it displays to me means that there is no WU which are actually suited relatively to my computer (i.e. made for PS3).

Why would my CUDA device be recognized without having the good drivers?

How can I see my CUDA drivers?

right click on desktop, you should see nvidia control panel, if you do, then look at the bottom left hand cornor of the window and you should see: System Information.

the driver version number can be found in that screen.

Why would my CUDA device be recognized without having the good drivers?

Computers store information about themselves in standard locations, and programs can read any or all of this.

e.g. The OS reads certain parts of the information coded into the cpu, and stores this.
BOINC reads this info and uses it to determine what type of cpu the computer has, and those parts of it's instruction set(s) that the OS has read. The instruction set info is displayed against the computers of each user in the CPU type.

Some people have noted that this doesn't include everything, such as all of the SSE sets. This is because the OS being used doesn't recognise these parts, being written before the cpu was created.

Other parts of the computer can also be determined, just by reading the information in certain locations.
So, BOINC can read the info for your graphics card, and determine, for those cards that it knows about, if it has cuda or not.
It can also read the name and version of the software that the graphics card is using.

Different projects have applications written for different graphics cards, and may well require certain versions of certain drivers in order to work.
This is the prerogative of the projects, and people wanting to use cuda need to check the web site(s) of the project in question to see what the requirements are.

right click on desktop, you should see nvidia control panel, if you do, then look at the bottom left hand cornor of the window and you should see: System Information.

the driver version number can be found in that screen.

Thanks for your answers but I think that you did not understand me correctly: I haven't the problem under Windows XP Pro (and I know how to find the drivers using XP). The problem occurs only when using Debian!

Okay, I installed last NVIDIA drivers, last CUDA apps, everything went smooth, my CUDA device is still discovered, named, but not used. Any news? Any project which offers CUDA services under Linux?

Thanks.

I'm having the exact same problem with Ubuntu 8.10 and nVidia driver 185.18.14. I made a symlink from the boinc home folder to libcudart.so in the cuda lib directory. I installed the cuda sdk apps and they all work showing "success". The cuda version I'm using is 2.2 which is what is listed for Ubuntu 8.10 on the nVidia CUDA download page. I tried nVidia driver 190 briefly, but there was no difference.

2.2 and 2.3 of cuda are very different, we've seen a 20+% increase on Seti.

cc_config, use gpus 1.


http://boinc.berkeley.edu/wiki/Client_configuration

<use_all_gpus>1</use_all_gpus>

I made a symlink from the boinc home folder to libcudart.so in the cuda lib directory.

Did the install runtime asked you to?

I installed the cuda sdk apps[...]

Which ones?

cc_config, use gpus 1.

<use_all_gpus>1</use_all_gpus>

What is its default value?

By default there is no value, as there is no cc_config.xml when you install BOINC. But the client only uses all but the first GPU detected. So essentially that value is zero, as in <use_all_gpus>0</use_all_gpus>

---

By default there is no value, as there is no cc_config.xml when you install BOINC. But the client only uses all but the first GPU detected. So essentially that value is zero, as in <use_all_gpus>0</use_all_gpus>

Thanks.