I am in the process of moving some workload over to Einstein@Home. Are there any optimized apps or configuration change I should be making?
Hello Aaron welcome. It looks like one of your hosts may have a problem with one of its GTX-780 cards. Host 1227521 error tasks
I had a look over a couple of the logs and they show the same card CUDA device #1 "GeForce GTX 780" generating the errors.
Quote:
[21:12:54][5456][INFO ] Starting data processing...
[21:12:54][5456][INFO ] CUDA global memory status (initial GPU state, including context):
------> Used in total: 567 MB (2507 MB free / 3074 MB total) -> Used by this application (assuming a single GPU task): 0 MB
[21:12:54][5456][INFO ] Using CUDA device #1 "GeForce GTX 780" (2304 CUDA cores / 822.22 GFLOPS)
[21:12:54][5456][INFO ] Version of installed CUDA driver: 8000
[21:12:54][5456][INFO ] Version of CUDA driver API used: 3020
Quote:
[22:22:10][5456][ERROR] Error during CUDA host->device HS thresholds data transfer (error: 719)
[22:22:10][5456][ERROR] Demodulation failed (error: 1007)!
You'll need to look into this, if that cards is over-clocked i would return it to standard settings and see if the problem continues.
That computer has all the cards set to stock clock speeds. So I am not sure what the problem may be. They are all running 2 jobs a card.
I could not find an example but has that card completed task ok yet?
I would try
* running a single task per card just to see if that solved the problem. This is not very efficient.
* cuda5.5 as mentioned by Zalster and see if the problem goes away.
You might consider
* a different driver version 368.22 was released recently - or step back a few releases.
Just moved over from SETI@home, what do I need to optimize?
)
On your home page, under Einstein prefences (default) setting the 5th click about is "run test applications"
you want to click that
It will allow you to Run the Parkes Cuda 55 which is a "test" app, even though we have been using it for many months now.
It runs faster than the traditional Parkes cuda 32 on Nvidia cards
You can also try changing the GPU utilization factor but I'll let someone else explain that part.
I personally just use a app_config so....
Awesome, thanks a
)
Awesome, thanks a bunch!
All the best,
Aaron Lephart
RE: I am in the process of
)
Hello Aaron welcome. It looks like one of your hosts may have a problem with one of its GTX-780 cards. Host 1227521 error tasks
I had a look over a couple of the logs and they show the same card CUDA device #1 "GeForce GTX 780" generating the errors.
You'll need to look into this, if that cards is over-clocked i would return it to standard settings and see if the problem continues.
Good luck.
BTW I have my 750ti
)
BTW I have my 750ti configured as such:
einsteinbinary_BRP6
.5
.20
All the best,
Aaron Lephart
Thanks, I actually became
)
Thanks, I actually became aware of that card erroring out.
That computer has all the cards set to stock clock speeds. So I am not sure what the problem may be. They are all running 2 jobs a card.
All the best,
Aaron Lephart
RE: That computer has all
)
I could not find an example but has that card completed task ok yet?
I would try
* running a single task per card just to see if that solved the problem. This is not very efficient.
* cuda5.5 as mentioned by Zalster and see if the problem goes away.
You might consider
* a different driver version 368.22 was released recently - or step back a few releases.
Good luck