TThrottle is not throttling Milkyway@Home gpu tasks on 3.52

[book_wyrm]

I joined the Milkyway@Home project today it queued about 10 gpu tasks for my Nvidia 260GTX, driver 270.61 on Win7 64.
After the first task finished and second task started I noticed that the temp of my gpu was increasing way over the limit that I had set on TThrottle.
TThrottle was stating that it was throttling the gpu at 2%
The only solution I could find was to restart TThrottle. Even the general throttling in the rules tab had no effect.
After this second task finished, the same thing happened with the third task and so on. Since the tasks last just over 25mins, it would be a pain to manually restart TThrottle each time a task finishes.

At first I assumed that TThrottle wasn't aware of the first task finishing and the next starting, but the PID in TThrottle was the same as the PID in task manager and the task name was the same as BOINC's.
TThrottle works fine with GPUgrid on this PC

[Pepo]

TThrottle was stating that it was throttling the gpu at 2%
The only solution I could find was to restart TThrottle. [...] After this second task finished, the same thing happened with the third task and so on. Since the tasks last just over 25mins, it would be a pain to manually restart TThrottle each time a task finishes.

You did not state whether restarting TThrottle helped at all to lower the GPU usage and temperature?

Additionally, it could also be possible, that even these 2% of available CPU time is enough for the application, to utilize the GPU that much. I can see similar effects with Collatz and PrimeGrid. Even with very strong throttling, the GPU is still being loaded pretty much.

[book_wyrm]

Restarting TThrottle does drop the temperature down to the limit I've set with gpu utilization at about 55% or so for each Milkyway@Home gpu task.

[Pepo]

I joined the Milkyway@Home project today it queued about 10 gpu tasks for my Nvidia 260GTX, driver 270.61 on Win7 64.

I've read some negative quotes on v.270 on BOINC fora, but am not sure, whether it might be relevant - rather not I suppose. (I do not want to suggest you to downgrade and try. I'm just waiting on another version.)

[fred]

Are you sure there isn't any other program that may cause this.
You can try the following, go to the BOINC tab and remove the check at "Connect with BOINC client"
See if the GPU entry is different. Other programs, different PID.

And I would like the BoincTasks or BOINC Manager entry for that task. I like to see the Plan Class an Use.

[book_wyrm]

Disabling the "Connect with BOINC client" seems to have fixed my problem.
After 1 Milkyway@home GPU task finishes the PID changes to the application of the next task.

The names of the tasks that are causing trouble all start with:
de_separation_10_3s_fix20_1_XXXXXXX_XXXXXXXXXX_X such as de_separation_10_3s_fix20_1_1000505_1303917207_1
de_separation_13_3s_free_1_XXXXXXX_XXXXXXXXXX_X such as de_separation_13_3s_free_1_1164154_1303938523_0

I seem to get about a dozen tasks from Milkyway at a time so I'll keep checking each time it switches task to see if the temp increases too much.
I've been using TThrottle with GPUGrid for the past 5 months without problems so I'd be interested to see whether this is a BOINC, Milkyway@Home or TThrottle issue.

Lastly, what do you mean by "Plan Class an Use"?

[fred]

Disabling the "Connect with BOINC client" seems to have fixed my problem.
Lastly, what do you mean by "Plan Class an Use"?

(Plan Class)  = (cuda)                                                                                                                                        Use = (0.04C + 1NV)
SETI@home   6.08 setiathome_enhanced (cuda)   22fe11ac.786.14791.14.10.242_1   11:51:45 (00:09:48)   1.38   100.000   -   30-05-2011 15:18   0.04C + 1NV         Ready to report   

[book_wyrm]

Another Milkyway@Home task has finished for me and everything looks normal when the next task started.
This is with "Connect with BOINC Client" turned off.

plan class=(cuda_opencl) and Use=0.05C+1NV  for these tasks

[fred]

plan class=(cuda_opencl) and Use=0.05C+1NV  for these tasks

Thanks I will investigate.

[fred]

plan class=(cuda_opencl) and Use=0.05C+1NV  for these tasks

When possible, give me the line of the running tasks in the first TThrottle tab like:
CPU:0, GPU:1, PID:4740 (5)   Slot:0   http://setiathome.berkeley.edu/   22fe11ac.4712.14791.15.10.218
Gpu: lunatics_x32f_win32_cuda30_preview.exe, PID: 4740, Threads: 5

Please add 2 entries with connect with "BOINC with client" checked and unchecked.

[book_wyrm]

Just some clarification. Do you want 2 entries of separate tasks with "Connect with BOINC Client" off and 2 entries with it on?
Also do you want both entries to come from the same "instance" of TThrottle or restart between checking the "Connect with BOINC Client" setting?

Just blindly copying your example I have:
CPU:0, GPU:1, PID:4592 (5)   Slot:3   http://milkyway.cs.rpi.edu/milkyway/   de_separation_13_3s_free_1_1164206_1303938523
Gpu: milkyway_0.52_windows_intelx86__cuda_opencl.exe, PID: 4592, Threads: 5

[fred]

Just some clarification. Do you want 2 entries of separate tasks with "Connect with BOINC Client" off and 2 entries with it on?
Also do you want both entries to come from the same "instance" of TThrottle or restart between checking the "Connect with BOINC Client" setting?

Just blindly copying your example I have:
CPU:0, GPU:1, PID:4592 (5)   Slot:3   http://milkyway.cs.rpi.edu/milkyway/   de_separation_13_3s_free_1_1164206_1303938523
Gpu: milkyway_0.52_windows_intelx86__cuda_opencl.exe, PID: 4592, Threads: 5

Rather the one that isn't throttled. , So I can see what's going on.
So with the check in place.
Next I want the entry for the GPU with the check still in place.
Then remove the check and give me the entry for the same task so I can compare them.
The sample above looks perfectly OK to me. Same PID same nr of Threads.
But it can take to up to a minute or so for the task to be recognized.

[book_wyrm]

CPU:0, GPU:1, PID:2196 (5)   Slot:3   http://milkyway.cs.rpi.edu/milkyway/   de_separation_13_3s_free_1_1164186_1303938523
Gpu: milkyway_0.52_windows_intelx86__cuda_opencl.exe, PID: 2196, Threads: 5

GPU was overheating quite a bit on this task at about 83C. GPU was using 2%. After I turned "Connect with BOINC Client" off it didn't change temperature at all.
Restart of TThrottle was needed to reduce temperature.

[book_wyrm]

Milkyway is my backup project so I've not had an WU from it for a while

The problem still persists, my GPU still overheats until I restart TThrottle. Turning off 'Connect with BOINC client' has no effect anymore so I need to restart TThrottle no matter what settings I have on it.

Here's another task that's overheating:

Gpu: milkyway_0.52_windows_intelx86__cuda_opencl.exe, PID: 4292, Threads: 5
CPU:0, GPU:1, PID:4292 (5)   Slot:1   http://milkyway.cs.rpi.edu/milkyway/   de_separation_10_3s_fix20_1_114925_1304776310

I can now view my tasks on their website and it turns out that one of my tasks took about 18.88 seconds CPU time while it took 3,840.49 in real time with TThrottle running with GPU usage at about 33%
Is this relevant?
Previously, Milkyway was deleting task entries from my account information on their website as they were completed so I could only see what tasks I had running but not tasks that had been completed, until now.

[book_wyrm]

The task in my last post took 3,131.24 seconds real time and 15.49 CPU time

CPU:0, GPU:1, PID:3332 (5)   Slot:1   http://milkyway.cs.rpi.edu/milkyway/   de_separation_10_3s_fix20_1_114924_1304776310
Gpu: milkyway_0.52_windows_intelx86__cuda_opencl.exe, PID: 3332, Threads: 5
This task initially had "Connect with BOINC client" ticked