V9 of modified SETI MB CUDA + opt AP package for full GPU+CPU utilization

[efmer (fred)]

No need to update - never break working system
If you added AP 5.03 - all fine. Only AMD team app speeded up a littlein new release, all other enhancements should decrease amount of cases when GPU stays idle.

The Ap is running now there is enough work. If your are interested I noticed a definite correlation between the cuda.exe and the GPU temperature and run time. The faster the cuda.exe runs (the more time it gets to run  or the CPU is faster e.g. overclock) the GPU does more work in less time. So maybe optimizing the cuda.exe run time is not a bad idea afterall.

What is cuda.exe  ??

A process like MB_6.08_mod_VLAR_kill_CUDA.exe

[|MatMan|]

I don't know if it was asked before: Can I use this mod to crunch on the GPU even with BOINC installed in protected (= service) mode?

[efmer (fred)]

I don't know if it was asked before: Can I use this mod to crunch on the GPU even with BOINC installed in protected (= service) mode?

If you don't use Vista yes.

[Jason G]

Hi Raistmer,
    I've had some feedback by PM that the v9 Cuda app locked up (In the installer package I'm making)

 Is it sufficient for me to replace the DLL's with the old ones from v8a?  Does this mean the v9 ones need recalling / rebuilding or just the packages updated & replaced?

Jason

[Geek@Play]

I now believe the cuda app is still running but in Boinc Manager it says it is not running.  The GPU temp is up and after a while the work unit completes and it goes on to the next one.  Boinc manager may display working again but after a while it again says it's not working.

Strange.

[Raistmer]

Hi Raistmer,
    I've had some feedback by PM that the v9 Cuda app locked up (In the installer package I'm making)

 Is it sufficient for me to replace the DLL's with the old ones from v8a?  Does this mean the v9 ones need recalling / rebuilding or just the packages updated & replaced?

Jason

Replacing the libs should work fine. V8 and V7 were build with CUDA 2.1 too but I took DLLs from stock app then... And now I decided to do "all correct" and took CUDA 2.1 RTL that came with used CUDA Tools... seems was wrong move.

[Raistmer]

I now believe the cuda app is still running but in Boinc Manager it says it is not running.  The GPU temp is up and after a while the work unit completes and it goes on to the next one.  Boinc manager may display working again but after a while it again says it's not working.

Strange.

I have read somewhere that BOINC works better if no one looking on it Maybe it's the same case

[Jason G]

Replacing the libs should work fine. V8 and V7 were build with CUDA 2.1 too but I took DLLs from stock app then... And now I decided to do "all correct" and took CUDA 2.1 RTL that came with used CUDA Tools... seems was wrong move.

Bummer! ... Oh well .. better remove that installer then from the testing thread    Are you updating the  packages?  Perhaps we should test them in testing area first?

[Raistmer]

Replacing the libs should work fine. V8 and V7 were build with CUDA 2.1 too but I took DLLs from stock app then... And now I decided to do "all correct" and took CUDA 2.1 RTL that came with used CUDA Tools... seems was wrong move.

Bummer! ... Oh well .. better remove that installer then from the testing thread    Are you updating the  packages?  Perhaps we should test them in testing area first?

It worked just fine on my host while there were work.... They disabled servers even before my host uploaded all work done     
Didn't see any work for CUDA more 24 hours already 

[Raistmer]

Just for info:
there is
//limit to 8 GPU's for now
comment in CUDA MB source.
That is, current CUDA MB build can't support more than 8 GPUs per host.

[Gecko_R7]

Just for info:
there is
//limit to 8 GPU's for now
comment in CUDA MB source.
That is, current CUDA MB build can't support more than 8 GPUs per host.

Given Nvidia's issues @ 40nm process....might be a while yet before Quad-GPU on dual PCB.....so luckily not much need for a while.

However, bet we see this from AMD/ATI sooner than we think. 

[Geek@Play]

My four computers have been running fine the last 2 hours.  Perhaps the electrons just needed to be trained, I don't know.  I did download the 3 dll's directly from Seti (fanout) and will try replacing them if I have further problems.  Will advise here.

[edit]  Just after this post one hung up.  Copied in the stock dll's and rebooted.  It's running again.  Will advise.

[edit again] And another cuda app hung up.  Now installed stock app into all 4 machines, rebooted and running again.  Will adivse.

[Geek@Play]

No Good.

Still looks like the MB_6.08_mod_CUDA_V9.exe is hung up in Boinc Manager.  It shows "Waiting to run, .04 cpu's, 1 CUDA".  But the GPU temp is still up indicating that it is still running.

I have seen many times in the last hours that the work does complete and starts a new work unit.  But it can show as described above at any time.

Don

[Jason G]

is that with the older DLL's now?

[Geek@Play]

Yes.......now using the older dll's.  Also on two of my machines I have had to restart Boinc in order to get the CUDA app going again.  Looks like a hard lock up can also occur.