Have been running some "wall clock" timeing tests and clearly, I am better off restricting a 4cpu system to 3cpu when processing seti cuda even with the latest 6.6.2

For one second tic (both were running at high priority)

4cpu, high priority: .015% increment on progress
3cpu, high priority: .35 % increment on progress

That is 23:1 improvement and gives just under 3 seconds to do 1%.

Avg WU with 3 cpus: 5 minutes
Avg WU with 4 cpus: 115 minutes

gtx280, Q6700 cpu, vista 64

Similar finding (but not as great a savings) with a 9800gtx can be found here

Which project, which version of the project's science application, and which data file (workunit)?

From the screenshot in your last thread, you were working on SETI (main project) at that time. Again, at that time, the SETI project was issuing v6.06 of their CUDA-enabled application. In Beta testing, that version exhibited the same variation you've described between 3+1 and 4+1 operation.

Overnight, SETI have released v6.08 on the main project. I think you may find that it handles 4+1 rather better: but I don't know anyone who's performed that particular combination experiment (v6.2.2 + v6.08, 3+1 vs. 4+1) to confirm. When you get v6.08, it would be good to find out - but be careful: according to reports at SETI, the project messed up the checksum for the downloads of the new files, so new tasks are erroring out. With the new CUDA quotas, that could be a very big waste of bandwidth. Best to set NNT for the time being ...... oh.

The other variable in the equation (as if that little lot wasn't enough) is task variability. When you get v6.08, you'll be able - for the first time - to run SETI's "VLAR" tasks. But they run very slowly: more like your 4cpu figure. So it wouldn't be helpful to do the 3+1/4+1 test while a VLAR is going thropugh the works: wait for a more 'normal' task to test.

Which project, which version of the project's science application, and which data file (workunit)?

From the screenshot in your last thread, you were working on SETI (main project) at that time. Again, at that time, the SETI project was issuing v6.06 of their CUDA-enabled application. In Beta testing, that version exhibited the same variation you've described between 3+1 and 4+1 operation.

Overnight, SETI have released v6.08 on the main project. I think you may find that it handles 4+1 rather better: but I don't know anyone who's performed that particular combination experiment (v6.2.2 + v6.08, 3+1 vs. 4+1) to confirm. When you get v6.08, it would be good to find out - but be careful: according to reports at SETI, the project messed up the checksum for the downloads of the new files, so new tasks are erroring out. With the new CUDA quotas, that could be a very big waste of bandwidth. Best to set NNT for the time being ...... oh.

The other variable in the equation (as if that little lot wasn't enough) is task variability. When you get v6.08, you'll be able - for the first time - to run SETI's "VLAR" tasks. But they run very slowly: more like your 4cpu figure. So it wouldn't be helpful to do the 3+1/4+1 test while a VLAR is going thropugh the works: wait for a more 'normal' task to test.

The WU's were the 6.06 as shown here. Since last night seti has tried to send me 6.08 but the download is failing as shown here However, I am surprised that they even showed up as a failure as I have had "no new tasks" specified since late yesterday. I assume the download failed because of the checksum and not because the "no new tasks"

The WU's were the 6.06 as shown here. Since last night seti has tried to send me 6.08 but the download is failing as shown here However, I am surprised that they even showed up as a failure as I have had "no new tasks" specified since late yesterday. I assume the download failed because of the checksum and not because the "no new tasks"

Yes, SETI seem to have borked the v6.08 installation - Murphy supplied a faulty checksum. That's almost certainly the reason for recent SETI work errors.

Once SETI has released a new app, all new work will be tagged for computation with that app. So you'll be getting allocated the same tasks as before, but they will have that '6.08' designation beside them, and they will all fail - until you reach that daily quota.

If you're sticking with BOINC v6.2.2, I strongly suggest you disable network activity - perhaps a more reliable way to avoid these supernumerary work requests under current circumstances.

If you're sticking with BOINC v6.2.2, I strongly suggest you disable network activity - perhaps a more reliable way to avoid these supernumerary work requests under current circumstances.

I just discovered that even the project "suspend" in combination with "no new tasks" does not stop seti from attempting a download. At least only one download was attempted after I clicked on "update" By "only one download" I mean that there were 3 data type files eg: ...23, ...17, ...2 followed by the download of the 6.08 executable which then failed the checksum test.

01/21/09 09:00:08|SETI@home|Sending scheduler request: Requested by user.
01/21/09 09:00:08|SETI@home|CPU work request: 0.00 seconds, 0 instances
01/21/09 09:00:08|SETI@home|CUDA work request: 0.00 seconds, 0 instances
01/21/09 09:00:08|SETI@home|Reporting 2 completed tasks
01/21/09 09:00:14|SETI@home|Scheduler request completed: got 3 new tasks
01/21/09 09:00:16|SETI@home|Started download of setiathome_6.08_windows_intelx86__cuda.exe
01/21/09 09:00:16|SETI@home|Started download of 10dc08aa.12680.17250.14.8.23
01/21/09 09:00:31|SETI@home|Finished download of 10dc08aa.12680.17250.14.8.23
01/21/09 09:00:31|SETI@home|Started download of 10dc08aa.12680.17250.14.8.17
01/21/09 09:00:41|SETI@home|Finished download of 10dc08aa.12680.17250.14.8.17
01/21/09 09:00:41|SETI@home|Started download of 10dc08aa.12680.17250.14.8.2
01/21/09 09:00:50|SETI@home|Finished download of 10dc08aa.12680.17250.14.8.2
01/21/09 09:01:11|SETI@home|Finished download of setiathome_6.08_windows_intelx86__cuda.exe
01/21/09 09:01:11|SETI@home|[error] Signature verification failed for setiathome_6.08_windows_intelx86__cuda.exe
01/21/09 09:01:11|SETI@home|[error] Checksum or signature error for setiathome_6.08_windows_intelx86__cuda.exe

At least they didnt pop up pictures of "miss seti" faster then I could close the popups.

Finally got some 6.08 seti cuda and the results indicate there is not much difference if 3 or 4 cpu's are assigned. My calculations were based on watching a single WU increment its %progress bar while I timed it.

I ran two tests, one on a 9800gtx, vista 64 AMD, and the other a gtx280 Vista64 Intel

The gtx280 system could increment the %progress bar in about 10 seconds ie: from 20% to 21%

The 9800gtx took little over 12 seconds to do the same 1% increment.

The WU's were different of course, but I kept the % range near %20 as I noticed that some WU pick up speed the closer they get to the end.

The difference switching from 3 to 4 cpu was not significent like it was with the 6.06 cuda where there was a 23:1 improvement.