TJ, are you getting tasks consistently now, or as erroneous as before the driver update? The driver version really shouldn't cause such behaviour (unless the BOINC log said "driver version xxx too low, please upgrade to at least xxx").

MrS

Can't speak for TJ, but me I'm getting short tasks consistently now :)

---

Team Belgium

TJ, are you getting tasks consistently now, or as erroneous as before the driver update? The driver version really shouldn't cause such behaviour (unless the BOINC log said "driver version xxx too low, please upgrade to at least xxx").

MrS

Hello MrS, Yes I get tasks consistently. Directly after I did a clean install of the driver and a reboot I got them (cuda42). They take 25xxx to 26xxx seconds on the 550Ti. No BOINC didn't came with a driver this time (306.97 should do it as it ran beta's (cuda42) consistently.

Some off topic information:
On another pc with GTX285 I get consistently cuda31 WU's, they run 16xxx to 17xxx seconds. On the 550Ti they took 40xxx seconds.

---

Greetings from TJ

TJ, are you getting tasks consistently now, or as erroneous as before the driver update? The driver version really shouldn't cause such behaviour (unless the BOINC log said "driver version xxx too low, please upgrade to at least xxx").

MrS

Hello MrS, Yes I get tasks consistently. Directly after I did a clean install of the driver and a reboot I got them (cuda42). They take 25xxx to 26xxx seconds on the 550Ti. No BOINC didn't came with a driver this time (306.97 should do it as it ran beta's (cuda42) consistently.

My GTX 560 takes 19xxx to crunch a short 4.2 WU under Linux.

---

Team Belgium

Thanks for the updates, guys. So in this case it really seemed to be some weird driver issue!

MrS

---

Scanning for our furry friends since Jan 2002

Thanks for the updates, guys. So in this case it really seemed to be some weird driver issue!

MrS

Not on my side, though. I didn't change drivers at all but work started flowing after one of the scientists submitted 1000 WUs to the short 4.2 queue.

---

Team Belgium

Thanks for the updates, guys. So in this case it really seemed to be some weird driver issue!

MrS

Not on my side, though. I didn't change drivers at all but work started flowing after one of the scientists submitted 1000 WUs to the short 4.2 queue.

And it could be that at the same time I coincidentally did the driver update.

---

Greetings from TJ

Now that that is settled... we still aren't sure if this server (GPUgrid server) will refuse to send tasks due to the driver on the host being "unacceptable", however you want to define unacceptable. It appears Einstein's server is configured to do so but that doesn't mean this project's server is. Can anybody verify it or think of a quick and easy way to test it? I suppose one could install an unacceptable driver and see if they still receive tasks. In fact I would do such a test but I'm not sure which driver(s) might be deemed unacceptable and I don't want to flail about with guesses.

---

BOINC <<--- credit whores, pedants, alien hunters

Now that that is settled... we still aren't sure if this server (GPUgrid server) will refuse to send tasks due to the driver on the host being "unacceptable", however you want to define unacceptable. It appears Einstein's server is configured to do so but that doesn't mean this project's server is. Can anybody verify it or think of a quick and easy way to test it? I suppose one could install an unacceptable driver and see if they still receive tasks. In fact I would do such a test but I'm not sure which driver(s) might be deemed unacceptable and I don't want to flail about with guesses.

My i7, BOINC 7.0.28, GTX285 with driver 280.xx did not get beta's and short runs from CUDA 3.1 and 4.2.
BOINC came with the message that I must update my drivers to get new work.
Thus 280.xx driver is a good "test" one but in my case with a GTX285.

---

Greetings from TJ

How come one of the "short" queues on the server status page has 10,000+ WUs, but I can't seem to get any of them? I'm running the long WUs fine.

And, come to think of it, why are there two "short" queues? One for 3.1? So are there a bunch of WUs in the 3.1 queue that the server refuses to give me because I can run 4.2 WUs?

The server is configured to check the driver version against its CUDA support level, in order to avoid sending WUs to clients which couldn't execute the corresponding app. That should basically be a simple "if driver > xxx then send work" statement.

MrS

---

Scanning for our furry friends since Jan 2002

How come one of the "short" queues on the server status page has 10,000+ WUs, but I can't seem to get any of them? I'm running the long WUs fine.

And, come to think of it, why are there two "short" queues? One for 3.1? So are there a bunch of WUs in the 3.1 queue that the server refuses to give me because I can run 4.2 WUs?

Indeed top is cuda31 and the second is cuda42. This is the same order as on the compute preferences settings. So the 10000+ are the cuda42 ones.
The server can see (I think) for which WU the GPU is best. I get only long runs on my GTX285 when I disable all the others.

Edit; MrS posted when I was typing...
Edit 2, Thanks MrS.

---

Greetings from TJ

How come one of the "short" queues on the server status page has 10,000+ WUs, but I can't seem to get any of them? I'm running the long WUs fine.

And, come to think of it, why are there two "short" queues? One for 3.1? So are there a bunch of WUs in the 3.1 queue that the server refuses to give me because I can run 4.2 WUs?

Indeed top is cuda31 and the second is cuda42. This is the same order as on the compute preferences settings. So the 10000+ are the cuda42 ones.
The server can see (I think) for which WU the GPU is best. I get only long runs on my GTX285 when I disable all the others.

Edit; MrS posted when I was typing...
Edit 2, Thanks MrS.

Are you sure? According to Nate in the News forum, the top one is the 4.2 queue

http://www.gpugrid.net/forum_thread.php?id=3298&nowrap=true#28658

---

Team Belgium

So, were out of long runs now? Any idea when we'll be getting more?

We seem to have dried up on short WUs. Can we get some more? Thanks

---

Team Belgium

Perhaps there is something not quite good with the server. I got these messages from time to time (when I was not watching the pc):
26/02/2013 16:03:06 GPUGRID Sending scheduler request: Requested by project.
26/02/2013 16:03:06 GPUGRID Requesting new tasks for CPU and GPU
26/02/2013 16:03:08 GPUGRID Scheduler request completed: got 0 new tasks
26/02/2013 16:03:08 GPUGRID Message from server: No tasks sent
26/02/2013 16:03:08 GPUGRID Message from server: No tasks are available for Short runs (2-3 hours on fastest card)
26/02/2013 16:03:08 GPUGRID Message from server: No tasks are available for ACEMD beta version
26/02/2013 16:03:08 GPUGRID Message from server: No tasks are available for the applications you have selected.

However, when I request manually by clicking the "update" button I almost immediately get new work.
Perhaps more people have this "issue"?

---

Greetings from TJ

Are you sure? According to Nate in the News forum, the top one is the 4.2 queue

http://www.gpugrid.net/forum_thread.php?id=3298&nowrap=true#28658

Pretty sure as it is mentioned in several posts that the cuda42 is faster then the cuda31. And looking at the server state page and the time WU's are sent back, the fastest must be the cuda42. Or I might have missed something.

---

Greetings from TJ

Are you sure? According to Nate in the News forum, the top one is the 4.2 queue

http://www.gpugrid.net/forum_thread.php?id=3298&nowrap=true#28658

Pretty sure as it is mentioned in several posts that the cuda42 is faster then the cuda31. And looking at the server state page and the time WU's are sent back, the fastest must be the cuda42. Or I might have missed something.

I wasn't talking about which is fastest, rather about which queue is which. According to Nate, the 4.2 queue is the top one in Server Status.

---

Team Belgium

This is all I get:

GPUGRID 02-26-13 13:54 Sending scheduler request: Requested by user.
GPUGRID 02-26-13 13:54 Requesting new tasks for NVIDIA
GPUGRID 02-26-13 13:55 Scheduler request completed: got 0 new tasks
GPUGRID 02-26-13 13:55 No tasks sent
GPUGRID 02-26-13 13:55 No tasks are available for Long runs (8-12 hours on fastest card)

Sure, it can say "no tasks available for long runs" in Event Log and then 5 minutes later you receive a long run. One explanation is that there are in fact none available when you request one but a few minutes later someone returns an errored result so then a resend is made available. Or there are none available then someone returns a success result which allows the server to generate a new task based on that success result (since they generate some tasks "on the fly", IIUC).

---

BOINC <<--- credit whores, pedants, alien hunters

it again dried out :(

---

Team Belgium