Possible bad batch of faulty Gravity Wave tasks
3 Mar 2023 20:35:00 UTC
Topic 229172
(moderation:
Got a bunch of tasks on host 12788501, all of which errored out with:
einstein_O3MDF_1.00_x86_64-pc-linux-gnu__GW-opencl-nvidia: `--tlCompartments' requires `--FreqBand' to be divisible without rest by `--dFreq', is 7e-08
Suspending GW work for the time being.
Hi Richard, Thanks for the
)
Hi Richard,
Thanks for the heads-up. I presume you would have messaged Bernd directly as well?
Cheers,
Gary.
No, I was just on my way out
)
No, I was just on my way out of the door when I saw them. And I doubt he would want to intervene at this time of a Friday night!
I see the same thing on host 12808716, but that's protected from further downloads by the same preference change.
Richard Haselgrove
)
I'm getting the same error with:
These are all with the VelaJr tasks.
Ideas are not fixed, nor should they be; we live in model-dependent reality.
Richard Haselgrove
)
I've sent him a message so hopefully he might get it in the morning.
He may want to use remote access at least to stop sending faulty tasks. Otherwise nothing might happen until Monday.
Cheers,
Gary.
It's just the 1.03 GW GPU
)
It's just the 1.03 GW GPU tasks that are failing. I am running some 1.03 GW CPU tasks with no issues.
Yes, mine were GPU tasks too
)
Yes, mine were GPU tasks too (for NVidia), so it might be another application problem - I didn't have any CPU tasks for comparison. But from the error message I posted, it seemed possible that the search parameters were at fault. All possibilities will have to be explored.
Edit - I looked through some of my failed tasks on the original machine, but couldn't find any that had been reallocated to a CPU. They'd all been tried (and failed) by other users on GPUs, and then cancelled.
Richard Haselgrove wrote:Edit
)
O3MDF does not have any CPU apps, it would be very unusual if they suddenly became O3MD1 tasks.
.
Bernd cancelled the batch
)
Bernd cancelled the batch (see tech news)
It's been re-started now. I got one bad task on the first attempt, but the ones being issued now seem to be working.
Richard Haselgrove
)
thanks