Messages

Really strange problem over at Rosetta.  The normal way to add the project fails to get data unless the following is added to hosts at
c:\windows\system32\etc\hosts

128.95.160.156 boinc-files.bakerlab.org
128.95.160.156 bwsrv1.bakerlab.org

So not even all_projects_list.xml is sufficient to make that project work.

At least I did not have to detach / re-attach to fix the problem.

If running windows then check the event viewer. I am running 1.94 with 10 systems and silent restart disabled.  The only problem I have seen is when I accidentally select "All Computers" when I have messages or history on the display.  If I am fast enough I can select a PC else I either wait it out (takes to long) or try to close the app and restart it.

Possibly you have some setting contributing to the problem or I am not using some feature that you are having a problem with.

Need additional HWINFO GPUs slots. I have six of the P104-100 (gtx-1070 equivalent) that are not recognized.
I set hwinfo32 to use 1..4 and set gpu-z to use #5 but there is no way to add the 6th.
I edited tthrottle.xml setting
<GPU_SETUP>
  <nvidia_number_gpu_phys>6</nvidia_number_gpu_phys>
<GPU_SETUP>

but it did not help.  GPU-z does not handle more than 1 temp. I tried running multiple gpu-z but tthrottle only saw 1 temp, not 2 on gpuz.

I moved this from BT wish list to tthrottle.

I ended up using OpenSSH for Win11 and the below command to transfer the updated file to my 'Shire2' PC.

Code Select Expand

scp C:\ProgramData\BOINC\all_projects_list.xml Shire2:C:\ProgramData\BOINC\all_projects_list.xml
I only did this as I have 12 systems and two days ago I had to scramble to log in remotely and shut down each one due to an electrical storm.  After I got: 

Code Select Expand

ssh Shire2 "shutdown /s /f /t 0" to work I tried that scp command which I used to update the gpugrid URLs.

I have 6 P102-100 cards, these are equivalent to GTX-1080Ti, but are in the $40 USD range.
They can be force installed in Win11 as P104-100 or using an unsigned driver as P102-100.  They are  not recognized by tthrottle except using GPUz or HWinfo32

1 - GPU-z (looking at Tthrottle display)
I ran 4 instances of GPU-z and set each instance to a different GPU. Tthrottle shows 4 GPUs in the external tab but only one temperature (the first instance) was available.  Possibly there is a setting to force 4 GPUs in tthrottle.xml but I do not know how to configure it.

2 - HWinfo (looking at Tthrottle display)
HWinfo Pro shows all 4 GPU temperatures in Tthrottle but only good for 12 hours.  The temperatures matched that revealed by the nvidia-smi app.  I uninstalled Pro and put in HWinfo32 and there was a notice that not all items were uninstalled.  TThrottle's GPU2 showed 1100c using HWinfo32.  All the other temps in TThrottle were ok and matched nvidia-smi.  BT reported the 4 temperatures but put in 0c for the problem 1100c temperature.

I brought up GPU-z and set instance to the GPU2 and got the correct temperature (in Tt).  In Tthrottle, I was going to disable GPU2 in HWinfo and enable it for GPUz.  I noticed that GPU2 in HWinfo now had the correct temperature. I closed GPUz and the temperature momentarily went back to 1100c.  While looking this strange problem the temperature toggled between the correct temperature and the 1100c temp.  Eventually everything in tthrottle "external" stabilized to the correct temperatures that matched nvidia-smi.

I am guessing that the left over setting from HWinfo64 Pro caused the problem in HWinfo32.  If I had known the uninstall would not be clean I would have used REVO to do the uninstall.

It would be nice to avoid using GPUZ or HWinfo.





[edit] I think I found the problem with HWinfo32:  While I had uninstalled HWinfo64 Pro, I had failed to close it and the uninstall did not bother terminating it.  I was running both the 32 and the 64 bit version of the app at the same time.  Probably was a coincidence the problem became fixed when running GPUz.  I have since rebooted and all is working OK with HWinfo32

With some projects it's a mess.
BOINC should be able to handle this, I think this is a BOINC problem.

I posted the same question over at BOINC first and so far my question received 1300+ views here and less than 90 over at BOINC. There is clearly a BOINC problem.

This must be a windows problem. I am running 24h2 and while the popups do go originally to monitor 1. If I move the popup to monitor 2 it remembers to pop up in monitor 2.  I have to do this for every popup, but once it gets moved it remembers the same location even if BT is moved back to monitor 1.  I only tested the boinctasks and boinc setting's popups.  If you restart windows with the displays left on the screen then windows must do the restore.  It would be convenient to have the popup always be on the same monitor.

I do remember back in win10 if my monitor 2 died there were some applications that were ghosted to monitor 2 and it was a real PITA to get them to move back to monitor 1.  Currently with win11 when my monitor 2 dies (it has a problem) all the icons and apps move to monitor 1 but they will return once I unplug and plug the monitor back in.  That did not happen in 10 and on occasion the ghosted apps remained ghosted even after the monitor was fixed.  By "ghosted" I mean that you see the apps icon in the task bar of both monitors but the app does now show up anywhere when you click on its icon. 

Windows 11 reboots roughly every 4 month if left to itself.  I assume you are running 10.

!!! Yea I missed that.  it is actually a BOINC type setting. I was looking in boinctasks.

I appears to work exactly the same way as the BOINC Manager's "Exclusive Applications" which I never use.  I had only one application that needed exclusive use of the GPU:  "DVDFab" which converted bluray disks to MP4.  With redbox and netflix no longer in the rental business I rarely use that app.

I attempted to put in an exclusive application for a remote computer using Boinctasks.  Firstly, what on earth are the three fields for?  There is only one piece of information Boinc needs - the name of the executable.  Secondly, when I browse for the file, it uses the folders on my local machine, not the remote one.  Thirdly, it didn't save it (this may be because I don't know which field to put what in, they're not labelled!)

I cannot find that option the way you are describing.  I looked at both windows versions of BT (the .js and the .exe)
I have been using the cc_config.xml editor in the .exe version to add (for example)

Code Select Expand

   <options>
      <use_all_gpus>1</use_all_gpus>
      <exclusive_gpu_app>DVDFab.exe</exclusive_gpu_app>
      <allow_remote_gui_rpc>1</allow_remote_gui_rpc>
   </options>

I am running older versions of boinc and just a few days ago looked at boinctasks.js for the first time. I have only windows system to run boinctasks on so I might be overlooking where those options are.

BT uses the local PCs all_projects_list.xml when adding a project.

Would it be possible to optionally use the all_projects_list.xml that is on the PC running BoincTasks?  That way I only need to update one file instead of 12 or more files.

ideally, the project would notify BOINC admin of the URL change which could then be propagated to users.  I do not think they inform anyone other than popping up 100's of notices about the change to the user.

I was able to add the project by overriding the default URL


<project>
<name>GPUGrid.net</name>
<id>6</id>
<url>https://www.gpugrid.net/</url>
<web_url>https://www.gpugrid.net/</web_url>

It should have been

<project>
<name>GPUGrid.net</name>
<id>6</id>
<url>https://www.gpugrid.net/gpugrid/</url>
<web_url>https://www.gpugrid.net/</web_url>


It would be nice if BOINC could handle this problem but looking around it is evident that some projects do not bother to inform BOINC admin what they are changing. 

[This project seems to have changed its URL. When convenient, remove the project, then add http://milkyway.cs.rpi.edu/milkyway]

The Berkeley supplied "all_projects_list.xml" seems to be outdated according to Milkyway, WCG, and possibly others.

For example the typical warning request (usual 100's of them)

This project seems to have changed its URL.  When convenient, remove the project, then add http://milkyway.cs.rpi.edu/milkyway

I did the requested change for several Milkyway and WCG systems and noticed that the ones I changed no longer show up when using the project selection side bar.  The projects that show the warning message are the only ones that show up in the filter. By "no longer show up" I mean the active tasks are not listed.

I edited the "all_projects_file.xml" on one of the problem systems and changed

<url>https://milkyway.cs.rpi.edu/milkyway/</url>         <web_url>https://milkyway.cs.rpi.edu/milkyway/</web_url>

to

<url>http://milkyway.cs.rpi.edu/milkyway/</url>         <web_url>http://milkyway.cs.rpi.edu/milkyway/</web_url>

but had no effect.  I assume the project sidebar is created by adding active project urls from each system in BT (I have 9) and one of the Milkyway entries has https and that one is found before the http one. Another problem I see is that the project's master url does not bother to use http even after making the required change from https to http like the project requested.

From looking at the Milkyway and WCG forum and their problems and lack of response (mainly Milkyway) I doubt if the projects is going to ask Berkeley to amend that "all_projects_list.xml" and some projects like wprop are not on the list anyway.  Wprop shows up in the project filter but the tasks are missing just like the Milkway and WCG tasks.

I am going to make a guess that if "https" and "http" are not included in the lookup the tasks will be found but that is just a guess.

I have a rule that suspends the GPU for 20 seconds then issues a resume.  I use it to restart a hung task.

Code Select Expand

"c:\Program Files\Boinc\boinccmd.exe" --host h110btc  --set_gpu_mode never
cmd.exe /c timeout.exe /T 20
"c:\Program Files\Boinc\boinccmd.exe" --host h110btc  --set_gpu_mode always
The rule works fine if Boinctasks runs when Windows starts which is the normal case.
It does not work if I run Boinctasks from the desktop shortcut.  I need to specify "run as administrator" else none of the 3 lines of code do anything.

It is inconvenient to have to "run as administrator".  Possibly there is some security setting or ownership change to allow the apps to work without having to run as administrator.  I tried a number of things but they did not help.

Those lines of code are in a script ".cmd" file.  It is not necessary to "run as administrator" if I chose to run the script manually.  However, Boinctasks needs to run as administrator to have the script file work.

I now believe this is a hardware problem of some type. The low CPU utilization reported is incorrect.  There is something different in the win11 OS causing this.

If replace the win11 boot disk with my win10 boot the CPU utilization is back to %100 as is normal.  I have the boinc apps on the D drive so whether I boot with 10 or 11 the BOINC apps pick up where they left off.

My CPU is "skylake-x" i9-7900x which has known heating problems and there are reports of "phantom throttling"

Recently I ran two CPU intensive apps.  The cpu utilization was %50 (or less) but they finished in the same time that a windows 10 comparable system finished.  I cannot run more than two or the system overheats I am considering removing the Intel heat spreader as is recommended by other skylake-x ownners.

I found the problem and the solution.  This problem only showed up in Win11 22H2.

I cannot start BOINCTASKS first.  I have to start BOINC and then BOINGTASKS.

If I start BT first it seems there is no thread or CPU switching and the throughput drops to 2x or 3x or worse depending on how many BOINC applications are running

Discussion over at Microsoft
https://answers.microsoft.com/en-us/windows/forum/all/terrible-performance-loss-with-win11-22h2-how-to/128d232a-6b23-4c44-b2e4-56649b590e73

[edit] Problem seems related to core isolation. If I disable core isolation the cpu utilization starts increasing and within a few minutes** is back up to %90 like it should be. If core isolation is enabled then ir-regardless of the order that boinc is started the cpu utilization drops and eventually the "red" wanning shows up.

**could take a while to rise if the work unit has been running at low utilization for a long time.  However, any new tasks quickly get to %90+

The problem is not the video board.  I just put the old video board back in and the problem is still there.
The problem is either the Dell bios upgrade from 206 -> 207 or the security changes to core and device protection enabled in windows 11 about the same time as that 3070 was put in.
In any event it is not caused by bointasks or tthrottle.

I went into the bios and disabled speed step and locked the CPU to the base frequency of 3.6ghz. CPU-z shows the CPUs running at 3.6.  There is no throttling and the work units for CPU bound tasks take twice as long as an older and slower Xeon.  It is as if the internal cache it not working anymore, or something is stealing the CPU cycles.

It is not permitted to downgrade the bios, will have to get with Dell support.

The %99 cpu and the almost twice as fast speed is the older Xeon. I would never have noticed this problem if I was not running boinc tasks. Windows 11 performance and resource monitoring do not show any problems.

[/img][/quote]