MB_6.08_CUDA_V12_ VLARKill_FPLim2048.exe Excessive CPU Usage

[rTx]

Running on a twin core AMD Athlon 64x2 5000+ 4Gb RAM, with a GeForce 9500 GT 512Mb, I get 50% CPU usage by MB_6.08_CUDA_V12_VLARKill_FPLim2048.exe - that is one core is fully occupied.

I am using Windows XP 32 bit fully updated (NVidia driver 6.14.11.8250).  BOINC 6.6.36 running as daemon.  I used the Unified Installer (0.2) a couple of days ago after I put the new graphics card into this machine (which was previously non-CUDA) to create the app_info.xml.  I had one previously which I had done myself.

I should add BOINC Manager reports only 0.04 CPU usage.  I am getting the 50% usage figure from Task Manager. I am running CPDN (100), Rosetta (100), WCG (100) but WCG project is suspended on this computer, and SETI (1). Figures in parentheses are weightings.  SETI is set in account preferences not to use CPU (just GPU).

As you would expect, if I suspend SETI project, I get about 49% each CPU Usage on the CPDN and Rosetta projects (up from about 24% each with SETI enabled)

Any suggestions, please?

[sunu]

I am using Windows XP 32 bit fully updated (NVidia driver 6.14.11.8250).  BOINC 6.6.36 running as daemon.  I used the Unified Installer (0.2) a couple of days ago after I put the new graphics card into this machine (which was previously non-CUDA) to create the app_info.xml.  I had one previously which I had done myself.

Not fully updated... You say you're running nvidia driver 182.50. I don't remember exactly, but I think our windows cuda client needs a later driver (185?). Anyway the latest NVIDIA driver is 190.38 http://www.nvidia.com/object/winxp_190.38_whql.html Try it. And if you haven't done already, get the latest 2.3 cuda libraries.

[rTx]

Thanks.  So I updated the Nvidia driver to 6.14.11.9038.  I had previously updated using the Windows update facility - I should have known it would be five months out of date!  CPU usage for the process is now down to about 1%.  So that fixed that.  I'll do the libraries next.

[sunu]

I'll do the libraries next.

Cuda 2.3 will give you a nice speedup.

[rTx]

Cuda 2.3 will give you a nice speedup.

Very much faster. Thanks.

[Josef W. Segur]

The title from the original post was causing the "Recent Topics" at the left of every page to be rather wide, I inserted a space to allow it to wrap. No criticism of the poster is implied, it's just a limitation of the site code.
                                                                            Joe

[alz]

Using MB_6.08_CUDA_V12_VLARKill_FPLim2048.exe I too have noticed some sets of workunits taking up the entire CPU, and taking over an hour to process.  Several hit my scheduler today.   This was on a Win2008 server running the latest NVidia drivers and 2.3 DLLs and using a GTX275.  The machine ID (for user Al Stein) is 5019936 and the workunit is 24au09ab.16409.25021.9.10.134_1_0.  (This unit should be completing within the next 10 minutes or so on that machine.)  I aborted the other units that began with 24au09ab.16409.25021.9.10.* since they were taking up too much CPU time, but let this one go to completion.

Any suggestions?

[Raistmer]

IF you see high CPU usage - look in stderr. If questions will remain, post stderr here.

[Josef W. Segur]

IF you see high CPU usage - look in stderr. If questions will remain, post stderr here.

resultid=1354433352 turns out to be one of those cases where the app thinks it has been told to run on an emulated rather than a real GPU, so it falls back to CPU. AFAIK there's no way to know how things got that way, for a dedicated crunching system a reboot might be recommended. Copy of the task details page attached.
                                                                Joe

[attachment deleted by admin]

[Raistmer]

Card removed or monitor unplugged...

[alz]

Thanks for the inputs.  The machine is in the office and I won't be able to check it out or reboot it until Monday - if I reboot from home it currently does not fully auto-signin.   The 275 has a dummy vga plug and the main monitor is on a 6200.  A similar setup is working fine with a 285 and dummy vga, same OS and all.  Maybe my home-built dummy vga needs replacing; will take another dummy vga to the office and get back to you on Monday.  Thanks!

[alz]

I checked the machine out today - no issue with the video and vga dummy as far as I could tell.  A reboot started getting the queued CUDA workunits running normally (no longer emulating.)  I guess it will be a mystery until the next time this happens.  I have two other machines like this, with vga dummies running Win2008 and the main monitor on a 6200.  Time will tell I guess.  If this machine does it again, I will swap the vga dummy.