News:

Follow BoincTasks on Twitter Facebook        Visit our website here.
BoincTasks cloud login is working again

Main Menu

Testing Version 3.52

Started by fred, April 23, 2011, 06:52:07 PM

Previous topic - Next topic

0 Members and 2 Guests are viewing this topic.

fred

Changed: Signed with new certificate 22-06-2013 (setup, exe, dll).
Fixed: AMD/ATI driver sometimes finds a NVidia GPU, in this case no GPU temperature is  shown in the graphic

Beyond

Quote from: fred on April 23, 2011, 06:52:07 PM
Fixed: AMD/ATI driver sometimes finds a NVidia GPU, in this case no GPU temperature is  shown in the graphic
Solved a problem where 1 of my XP64 machines was showing 3 GPUs (the third at 0 degrees C).  It actually has 1 ATI & 1 NVidia.  Now it's reporting accurately.

fred

Quote from: Beyond on April 24, 2011, 01:21:36 AM
Quote from: fred on April 23, 2011, 06:52:07 PM
Fixed: AMD/ATI driver sometimes finds a NVidia GPU, in this case no GPU temperature is  shown in the graphic
Solved a problem where 1 of my XP64 machines was showing 3 GPUs (the third at 0 degrees C).  It actually has 1 ATI & 1 NVidia.  Now it's reporting accurately.
Nice.
But why the ATI interface is detecting a NVidia GPU at all  :o

Beyond

Quote from: fred on April 24, 2011, 03:02:12 AM
Quote from: Beyond on April 24, 2011, 01:21:36 AM
Quote from: fred on April 23, 2011, 06:52:07 PM
Fixed: AMD/ATI driver sometimes finds a NVidia GPU, in this case no GPU temperature is  shown in the graphic
Solved a problem where 1 of my XP64 machines was showing 3 GPUs (the third at 0 degrees C).  It actually has 1 ATI & 1 NVidia.  Now it's reporting accurately.
Nice.  But why the ATI interface is detecting a NVidia GPU at all  :o
In my case it was on a dual boot machine.  In Win7-64 TThrottle detected properly.  In WinXP-64 it detected an additional GPU that wasn't there.

Pepo

#4
I was running two Collatz tasks (1.00C, 0.01C+1NV) pretty deeply throttled (GPU run time 2%, CPU 3%). I wanted to give them a free run for a while, so I've unchecked "Auto active" (otherwise, "Connected to BOINC is checked). (I've possibly  checked and unchecked "Auto active" one more time.)

After running with pretty loud fan for 10 minutes (2 CPU graphs were 10°C over threshold line, GPU graph 15°C over threshold line), both tasks were 3 times suspended ("Suspending computation - CPU is busy", my tasks are kept in memory) and resumed 3 times - 20 sec. suspend, 20 sec. run, 10 sec. suspend, 20 sec. run, 10 sec. suspend, then free run. (TimeStamp1)

Ten minutes later I've noticed that the fan is not loud anymore. Look at the graph - 90 seconds after TimeStamp1 the GPU temperature went down to around the regulation threshold line (by some -15 degrees), the CPU temperatures went down too (notebook - everything shares the same cooler) by some -5 degrees, but still stayed well over the regulation temperature. But "Auto active" was still unchecked and neither GPU nor CPU were snoozed. The GPU task was not checkpointing anymore (the last checkpoint was 8 seconds prior to (TimeStamp1)). Then I've checked the GPU task's process and noticed that its main thread is in suspended state. Checking and unchecking "Auto active" did not help - I had to resume the thread manually with Process Explorer. From that moment the GPU task started to compute, all temperatures jumped up...

(Yet again I was badly missing some 3 or 4 hours view mode in TTh's and BT's temp graphs - while writing this, I had to do a few phone calls and the events went over 60 minutes - the 12 hour window is already too coarse to recognize anything :'()

Could TTh leave one GPU task's thread in a suspended state, when leaving the "Auto active" state, or sometimes else? Maybe the "Auto active" checkbox state could be logged?


I've later tried to manually suspend the GPU task's main thread, while TTh was "inactive" - the same happened, GPU usage went to zero, the GPU temp settled to approx. 1 degree over my threshold. I could even suspend all its threads - the temperature did not go down. Just snoozing the GPU (or killing the process) caused the GPU temperature to go down. Could maybe someone enlighten me, whether the GPU tasks put GPU in some state, where it consumes much more energy, although they are obviously not able to read or write any data or commands?
Peter

fred

Quote from: Pepo on April 28, 2011, 09:26:15 AM

I've later tried to manually suspend the GPU task's main thread, while TTh was "inactive" - the same happened, GPU usage went to zero, the GPU temp settled to approx. 1 degree over my threshold. I could even suspend all its threads - the temperature did not go down. Just snoozing the GPU (or killing the process) caused the GPU temperature to go down. Could maybe someone enlighten me, whether the GPU tasks put GPU in some state, where it consumes much more energy, although they are obviously not able to read or write any data or commands?
The GPU is a bunch of independent CPU's, they run and may crash or do something unexpected.
It's possible that TThrottle leaves a Thread suspended, but only when there is a bug somewhere.