I will not use a modified driver. Please do not suggest that I do so.

I am also not in a position to buy a new graphics card simply to make SETI work.

Right now, my machine throws almost every work unit out as a computation error.

I see some comments here and there about newer stock drivers not working, but I do not see any clear and concise indications stickied at the top of this forum saying what we should be using.

Please sticky a thread with the most recent stock NVidia drivers that support SETI. even if it's not this thread. I'm confident that I'm not the only one that would appreciate an answer to this.

---

For some reason your system is trying to run the SoG application (which is the best at present), however your system only shows support for OpenCL 1.1, and 1.2 is the minimum requirement for SoG. As far as I can tell, the 500 series only offers support for OpenCL 1.1- no matter what the driver version, SoG will not run on that hardware.
If you run CUDA50 there should be no issues (from memory the GTX 560 series did hove some issues with very borderline hardware running on the ragged edge).

---

Grant
Darwin NT

Hrm. I thought BOINC detected hardware compatibility? If the only problem is that my hardware won't run certain work units, why am I getting that type of work units?

Also, is CUDA50 an NVidia version, or a BOINC version?

Just downgraded CUDA drivers to 5.0.3.

Now I'm getting an error message "NVidia had adequate OpenCL support for FERMI+ devices starting with 347.09 drivers. Please upgrade driver to 352 or higher to use OpenCL...

If the only problem is that my hardware won't run certain work units, why am I getting that type of work units?

The issue isn't the work unit, it's the application being used to process the WU.
As to why BOINC is trying to use that application for a non-supported OpenCL version, i've no idea.

Also, is CUDA50 an NVidia version, or a BOINC version?

It's the Application version used to process work.
Generally;
CUDA23: for Pre-Fermis ONLY - do NOT use on Fermi/Kepler/Maxwell
CUDA32: for mixed systems
CUDA42: for Fermis
CUDA50: for Kepler/Maxwell

---

Grant
Darwin NT

Just downgraded CUDA drivers to 5.0.3.

Now I'm getting an error message "NVidia had adequate OpenCL support for FERMI+ devices starting with 347.09 drivers. Please upgrade driver to 352 or higher to use OpenCL...

After downgrading and getting the message, I aborted the SOG tasks and got CUDA50 tasks. I will check tomorrow and see if that has fixed the issue.

I should be able to run CUDA50 tasks with the most recent NVidia drivers though, shouldn't I? It seems strange that SETI is trying to send me work units that my hardware simply cannot handle.

I should be able to run CUDA50 tasks with the most recent NVidia drivers though, shouldn't I?

Yep, while old it is the current CUDA application.
The SoG application based on OpenCL performs better, but if your hardware doesn't support it then CUDA is the way to go.

---

Grant
Darwin NT

I think some not so good advice is being handed out here.

With driver version 372.54 my GTX 550Ti's/560/560Ti were quite happy doing SoG tasks until I recently retired them.

In fact until you aborted all that SoG work the only error you had was a CPU task, your SoG tasks were completing (they're just waiting for your wingmen to validate them), check your task list, https://setiathome.berkeley.edu/results.php?hostid=7063850. ;-)

You just have to be sure that you reserve a CPU core to feed the SoG app (running CUDA tasks ATM is going backwards).

Cheers.

I think some not so good advice is being handed out here.

With driver version 372.54 my GTX 550Ti's/560/560Ti were quite happy doing SoG tasks until I recently retired them.

That driver gave then OpenCL 1.2 support? 3 of the results of my searches showed they were only good for 1.1

---

Grant
Darwin NT

I think some not so good advice is being handed out here.

With driver version 372.54 my GTX 550Ti's/560/560Ti were quite happy doing SoG tasks until I recently retired them.

That driver gave then OpenCL 1.2 support? 3 of the results of my searches showed they were only good for 1.1

As seen here they are OpenCL 1.2 capable with the right driver version (driver version 306.94 is too old and doesn't support OpenCL 1.2, but the one he had before did).

With my current BOINC version, 6.10.60, it doesn't have OpenCL detection at all, but it still works. ;-)

Cheers.

As seen here they are OpenCL 1.2 capable with the right driver version (driver version 306.94 is too old and doesn't support OpenCL 1.2, but the one he had before did).

I can't remember the driver version he had when I first looked at the system, but i'm pretty sure it was over the 306.94 minimum and it definitely showed only 1.1 support.

Ah, here we go from one of the std_err outputs.
Driver version: 372.90
Version: OpenCL 1.1 CUDA


Cheopis, ignore all before. I'm losing my mind.
That card, with the right driver, should support OpenCL 1.2 and therefore run SoG without problems.

---

Grant
Darwin NT

But then again here's what 1 of my SoG stderr output said with the 560.

 Name:						 GeForce GTX 560
  Vendor:					 NVIDIA Corporation
  Driver version:				 372.54
  Version:					 OpenCL 1.1 CUDA

And it worked error free for months with all my 5xx cards.

Cheers.

*slaps self in head*
I was reading the OpenCL AP, not the OpenCL MB, info.
*repeats slap*

---

Grant
Darwin NT

*slaps self in head*
I was reading the OpenCL AP, not the OpenCL MB, info.
*repeats slap*

There was no problems with running the AP app either. ;-)

Cheers.

With driver version 372.54 my GTX 550Ti's/560/560Ti were quite happy doing SoG tasks until I recently retired them.

In fact until you aborted all that SoG work the only error you had was a CPU task, your SoG tasks were completing (they're just waiting for your wingmen to validate them), check your task list, https://setiathome.berkeley.edu/results.php?hostid=7063850. ;-)

You just have to be sure that you reserve a CPU core to feed the SoG app (running CUDA tasks ATM is going backwards).

Cheers.

How do I reserve a core for GPU tasks? Never had to do that before.

With driver version 372.54 my GTX 550Ti's/560/560Ti were quite happy doing SoG tasks until I recently retired them.

In fact until you aborted all that SoG work the only error you had was a CPU task, your SoG tasks were completing (they're just waiting for your wingmen to validate them), check your task list, https://setiathome.berkeley.edu/results.php?hostid=7063850. ;-)

You just have to be sure that you reserve a CPU core to feed the SoG app (running CUDA tasks ATM is going backwards).

Cheers.

How do I reserve a core for GPU tasks? Never had to do that before.

In your BOINC Manager Preferences under the Processor Usage tab near the bottom is "On multiprocessor systems, use at most xxx.xx % of the processors" and change it to 87.5%.

You will also need to update that 306.94 driver back to the 1 you had installed previously, 306.94 is to old. ;-)

Cheers.