My computer ID 4697230 (Intel Quad Q6700) has now an NVIDIA GeForce 9600 GT GPU (CUDA capable, 512MB). Got the latest drivers from nVidia, and the latest BOINC software. After installing the card, the BOINC log reports that the CUDA driver was found, but the WU total processing time has not changed (still around one hr each). The computer still processes 4 WU's concurrently, still using optimizer "Windows optimized S@H Enhanced application by Alex Kan", and no signs that the GPU is actually taking over. Do I need to remove the optimizer? If so, will the GPU work faster than the 4 CPU's? Thanks!

---

You need a cc_config.xml file in your BOINC data directory. Copy the following to notepad and save as .xml

<cc_config>
<options>
<ncpus>5</ncpus>
</options>
</cc_config>

Remember to not save it as all files as it will try to stick .txt on the end. This goes in your BOINC DATA file NOT Seti at home.

---

PROUD MEMBER OF Team Starfire World BOINC

Or, since there's not that much work available from Seti anyway and there won't be until next week, why not wait for the release of BOINC 6.6 which is also scheduled for next week and which will fix this problem without you needing to use the client core configuration file. You will need to remove that file anyway when you upgrade to BOINC 6.6

---

Thank you for your reply. I added the cc_config.xml file as suggested, and now the pc is processesing 5 wu's, but none of them done by the GPU. Are CUDA wu's a different version of wu that are processed only by GPU's?
If so, I guess the temporary lack of new wu's also affects those CUDA wu's?
Thanks

---

Seti Enhanced CUDA is done by the 6.08 application, Seti Enhanced CPU work has the 6.03 application while Astropulse will use the 5.00 and 5.03 applications.

Any work already on your computer will already have the application appointed, so when you check your Tasks tab in BOINC Manager, Application column, you can easily see what is appointed to what. Adding the cc_config.xml file won't magically change the work already on your system to be appointed to another piece of hardware, you will have to download it appointed (or specified) for that application version.

---

Adding the cc_config.xml file won't magically change the work already on your system to be appointed to another piece of hardware, you will have to download it appointed (or specified) for that application version.

Learning a bunch here. So new wu's are appointed as soon as they are downloaded. Then how do I download a wu to be appointed to the GPU?

---

That's a bit touch and go here at Seti, since they only use one feeder. When you look at the bottom of the server status page, there is an explanation of what the various parts of the BOINC server back-end are and do.

But in short, BOINC on your computer has a client side scheduler and work fetch module. These will keep an eye on the amount of work you have and if you're in need for more work, the work fetch module will contact the Seti scheduler and ask for work.

The Seti scheduler will send your request through to the feeder, which will supply you with work. But because Seti only has one feeder, it depends on what is at the top of the stack at that time what kind of work you get. Even though they try to keep a 33 work units for each option available at all times (Astropulse, Multibeam CUDA, Multibeam CPU), it can happen that things run out. Especially now with the database problems.

With BOINC 6.4.5 to .7 the work request will be anything that you enabled and so you'll get anything that is available as it does not have a separate GPU scheduler yet. BOINC 6.6 will be able to specifically ask for work for the CPU or the GPU.

---

I have a similar problem,

SETI found my CUDA device an decided to fetch work for it. The point is, the WUs are appointed to SETI 6.08 (cuda), but the CUDA-device isn't doing any work. The CPU is crunching the WU for SETI and to other projects. How can this be?

---

I have a similar problem,

SETI found my CUDA device an decided to fetch work for it. The point is, the WUs are appointed to SETI 6.08 (cuda), but the CUDA-device isn't doing any work. The CPU is crunching the WU for SETI and to other projects. How can this be?

From your task 1191699082:

...
SETI@home using CUDA accelerated device GeForce 8600 GT
Restarted at 29.15 percent.
Cuda error 'cudaMalloc((void**) &dev_flagged' in file 'c:/sw/gpgpu/seti/seti_boinc/client/cuda/cudaAcceleration.cu' in line 358 : out of memory.
setiathome_CUDA: CUDA runtime ERROR in device memory allocation (Step 1 of 3). Falling back to HOST CPU processing...

Gruß,
Gundolf

---

Computer sind nicht alles im Leben. (Kleiner Scherz)

SETI@home classic workunits 3,758
SETI@home classic CPU time 66,520 hours

So this should tell me, that SETI ran out of memory on the graphic-card?
Interesting...

---

Could it be, that my settings produce conflicts?
I set BOINC to run only, if the computer is not used. Might those breaks in calculation cause this memory-error?

---

Could it be, that my settings produce conflicts?
I set BOINC to run only, if the computer is not used. Might those breaks in calculation cause this memory-error?

I had the same thing happening here on my GeForce 8600GT when I first started using BOINC on this computer - I would see memory allocation errors in my result files. Once I upgraded to the latest version of the NVidia drivers (don't recall the version right off but it's the newest one on their web site) this stopped happening and everything is running smooth as silk.

---

Could it be, that my settings produce conflicts?
I set BOINC to run only, if the computer is not used. Might those breaks in calculation cause this memory-error?

I had the same thing happening here on my GeForce 8600GT when I first started using BOINC on this computer - I would see memory allocation errors in my result files. Once I upgraded to the latest version of the NVidia drivers (don't recall the version right off but it's the newest one on their web site) this stopped happening and everything is running smooth as silk.

It didn't change anything, unfortunately...

---

Could it be, that my settings produce conflicts?
I set BOINC to run only, if the computer is not used. Might those breaks in calculation cause this memory-error?

I had the same thing happening here on my GeForce 8600GT when I first started using BOINC on this computer - I would see memory allocation errors in my result files. Once I upgraded to the latest version of the NVidia drivers (don't recall the version right off but it's the newest one on their web site) this stopped happening and everything is running smooth as silk.

It didn't change anything, unfortunately...

While I am by no means an expert on Vista (I use XP), I do remember reading several message saying that CUDA is incompatible with programs installed with the Protected Execution Application option (I think that's the correct name). Exactly what that is I don't know, as I have very little experience with Vista, but could this be your problem?

---

Just checked the configuration, the option is not set. So I will keep on searching...

---

Just checked the configuration, the option is not set. So I will keep on searching...

There's info in this message that might apply to your situation.

Edit: As well as this one and this one

---

Just checked the configuration, the option is not set. So I will keep on searching...

There's info in this message that might apply to your situation.

I think my problem is solved so far. I found some Vista Aero components, that still have been active. Due to the fact, that I need two monitors to work, Aero tool a large amount of memory. So far, CUDA works again... Thanks for the hint, Byron!

---

It's been several days now since my first post about not getting any CUDA units for my main CUDA-enabled PC, or no CUDA units were ever appointed. At first I thought that the fact that not too many new WU's were generated was the reason. Now I see that new WU's are available (none of my PC's ever ran out), but still...

I read in another posting that someone suggested to detach and reattach to SETI@Home; I tested that with another (slower) machine, and all the current WU's disappeared - I guess detaching means deleting of ALL WU's.

If I do the same with my CUDA PC, I would lose over 1000 WU's. Should I still wait? Should I detach/reattach? Is there any configuration that needs to change? Should I wait for Boinc 6.6?

---

I guess detaching means deleting of ALL WU's.

Uh yes, what else did you expect? Resetting the project and detaching from it will delete all work in queue, including finished tasks that are waiting to report. And no, there is no way for you to get them back.

Just wait. BOINC 6.6 is around the corner.

---

As a follow up, I upgraded to Boinc 6.6 several hours ago on my 4-core CPU, and although many new WU's were received, no appointment has been made to my GPU. Boinc still does report that a CUDA device was found, but all the WU's (old and new) are still setiathome_enhanced 5.28.
My web profile is set to use both CPU and GPU.
Is it possible that some leftover config from some older Boinc version is keeping it from using the GPU?
Should I remove the "Windows optimized S@H Enhanced application by Alex Kan" optimizer? I tried keeping and removing the cc_config.xml file where it tells the client core to use 5 processors.

---