My Seti task is 50% done in 2:05:03 with 5:13 remaining. Prsumably it will finish the remaining half in another 2 hours rather than the 5 hours indicated.

Is this the performance from CUDA?

David

---

My Seti task is 50% done in 2:05:03 with 5:13 remaining. Prsumably it will finish the remaining half in another 2 hours rather than the 5 hours indicated.

Is this the performance from CUDA?

David

It means the actual performance is probably better than the predicted performance.

This is one of the things that will tend to dial-in as the machine runs.

I don't want to blame this on CUDA. It just shows the difference between benchmarks and reality.

---

Is the code for the CUDA card up to date? I don't know what is correct and I am sure the thread owner would not know this as well.

---

My Seti task is 50% done in 2:05:03 with 5:13 remaining. Prsumably it will finish the remaining half in another 2 hours rather than the 5 hours indicated.

Is this the performance from CUDA?

I think Ned is right this time. It probably has more to do with BOINC having to dial-in the performance (or gauging it, if you will) of that particular machine. Once BOINC "learns" that machine, future work requests should be closer to reality so long as you run it as you normally would. If you purposely leave it on longer due to a deadline concern, BOINC will think that part of the norm.

---

My Seti task is 50% done in 2:05:03 with 5:13 remaining. Prsumably it will finish the remaining half in another 2 hours rather than the 5 hours indicated.

Is this the performance from CUDA?

David

Your computer have 2 CPU Cores + 1 GPU so you have to see
3 tasks running simultaneously (click on Progress column to sort the tasks).

If the task is setiathome_enhanced 6.03 it is computed by CPU
If the task is setiathome_enhanced 6.08 or 6.09 it is computed by GPU (Video card) (CUDA)

I will expect CUDA task to be computed in 30 minutes - not 4 hours

P.S.
One of yours CPU tasks completed (took 5.5 hours) and reported (still waiting for credit).
Still none of the CUDA GPU 6.08 tasks are reported as finished.

.

---

 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 

Still none of the CUDA GPU 6.08 tasks are reported as finished.

Probably they are all waiting to run because there's not enough available GPU memory.

Gruß,
Gundolf

---

Computer sind nicht alles im Leben. (Kleiner Scherz)

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

I only have two tasks running.

This might be important. Under messages from Einstein@home, "Message from server, your NVDIA GPU has insuicient memory (need 450MB)"

Seti message is"Message from server: No jobs available" That might be because I have already downloaded a couple hundred. They won't stop coming.

David

---

I only have two tasks running.

You should go to your SETI@home preferences and set

Use NVIDIA GPU if present no
(enforced by 6.10+ clients)

since your GPU doesn't have enough available memory. The client shouldn't download any more tasks in that condition, but obviously that bug is still there :-)

If you want to see the appropriate error message, you'll have to set up a cc_config.xml file and enable the <cpu_sched_debug> logging flag.

This might be important. Under messages from Einstein@home, "Message from server, your NVDIA GPU has insuicient memory (need 450MB)"

Einstein has other memory requirements than SETI.

Seti message is"Message from server: No jobs available" That might be because I have already downloaded a couple hundred.

Nope, those two are not related.

They won't stop coming.

That seems to be a bug in your BOINC client. See above and this thread, though the topic seems opposite.

Gruß,
Gundolf

Yes, it sounds as if he's got enough total GPU RAM to get Cuda work,
but then hasn't got enough available GPU RAM to actually start,
the 197.xx drivers did the same to my 9800GTX+,
it had enough RAM to get 6.09 work, but not enough to start it,

Claggy

That seems to be a bug in your BOINC client.

It is a bug in the present client(s) in the "- client: if a GPU job is blocked on available mem, don't fetch more jobs for that resource type." code. Changeset [trac]changeset:21197[/trac] has a possible fix, that will be seen in a future version of the client.

There's a caveat that when there are multiple GPUs in a system, with one (the first one) having run out of available video memory, whereas the rest have lots of memory free.

There's one task available, so BOINC will try to run work on the first GPU, find that it doesn't have enough memory and will then suppress work-fetch. This is because at this time work is assigned first, before the client checks available video memory. David will add code later on that will check available memory on GPUs when they're assigned.

---

Now I have three processes running. Beore there were only two running (not SETI). Now with 3 process running two are from the other projects and one is from SETI.

The SETI job has run 5:54:00 and is 21% complete. To completion shows 126 more hours!!

The other two jobs are running normally and will execute in the normal few hours each.

My computer is still happily downloading jobs from SETI but has only downloaded a handful of jobs from the other two projects(Einstein and Rosetta). I have engaged "no new tasks" on SETI so the downloading should stop. I probably have 200 jobs in the que here.

David

---

I have engaged "no new tasks" on SETI so the downloading should stop.

Unless you're one of those people who wants a 100% manually operated BOINC, I would strongly discourage setting "no new tasks."

It may be that SETI can't use your CUDA card, and setting your preferences on the web site would stop work from being assigned to it.

---

I have a similar, but probably not CUDA related problem as OSX does not appear to have any CUDA enabled apps. Anyway, my account lists >45 work-units as in-progress. All of these work units appear in my "/Library/Application Support/Bionic Data/projects/setiathome.berkely.edu" folder but are not being used. Bionic continues to download new work units. I'm sure that I can meet the deadlines for the work units if I can get them started soon.

Does anyone have any ideas how I can get SETI to run the jobs in my data folder?

I have a similar, but probably not CUDA related problem as OSX does not appear to have any CUDA enabled apps.

SETI has no stock application for OSX, but since you are running an optimized application, those could very well be CUDA tasks.

Does anyone have any ideas how I can get SETI to run the jobs in my data folder?

You'll have to make more GPU memory available. To see how much is missing, create a cc_config.xml file with the <cpu_sched_debug> logging flag set.

If you can't seem to run the CUDA tasks, disable using your GPU in the SETI@home preferences and abort the tasks, so that your wingmates won't have to wait for them to expire.

Gruß,
Gundolf

---

Computer sind nicht alles im Leben. (Kleiner Scherz)

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

Thanks for the suggestion.
However, I get my Mac OSX Core2 Duo optimized apps from http://www.arkayn.us/. These apps do not have cuda support. Regardless, I set my SETI@home preferences to not send GPU work.
I did get SETI@home to run the remaining jobs in the folder by setting 'no new jobs'. I still have a big backlog, but most of the new jobs are those strange 1hr jobs.

Thanks again,
Chris