Computation error - Float Invalid Operation!
14 Dec 2018 12:53:45 UTC
Topic 217441
(moderation:
I have been contributing to different BOINC-projects since 2006 and recently joined Einstein@home.
Now when I check my account / stats I have 39 Error out of 139 Valid. Im using about 4 machines for this project with intel cpu, intel gpu/nvidia gpu, win10. Most machines run 24/7, boinc installed as service so it can compute while not in use.
When checking Task info, for exampel https://einsteinathome.org/sv/task/810669552, they all give same type of error information.
Unhandled Exception Detected...
- Unhandled Exception Record -
Reason: Float Invalid Operation (0xc0000090) at address 0x00000000004ed8af
Engaging BOINC Windows Runtime Debugger...
After discovered the high error-rate to valid, I have now limited Einstein project priority until I know more.
Suggestions ?
There was some kind of
)
There was some kind of problem with some tasks on that gravitational search:
https://einsteinathome.org/goto/comment/168234
Bernd wrote that message December 13 and problem was fixed December 14. I see your computer received all those tasks December 12 at latest. Maybe them were from that broken batch.
You could limit the work quote to small value and let your computer receive a couple new gravitational tasks now. If they still crash then it's something else.
Your service install of Boinc
)
Your service install of Boinc will prevent you from using the GPUs for crunching as Windows uses it's own basic driver when a user is not logged in and Boinc can't detect your GPU then. And you should always use drivers from GPU manufacturers site and not what windows 10 tries to give you. Microsoft usually leaves the OpenCL component out of their driver package which is needed here.
I think that the floating
)
I think that the floating point invalid problem is back again. Starting with a download on 26/12/18 I got a solid bunch of 'error whilst computing' results. ( I only connect about once a day so the start time may be before that date). Some of the wu have not been reissued but quite a lot have and they have all failed on many other peoples computers.
The errors are on my computer 12469662.(Peter-PC)
A typical wu is 384558386. h1_0495.65_O1C02Cl2In0__O1OD1_495.90Hz_1463_0
This on a Gravitational Wave All-sky search on LIGO O1 Open Data v0.01 windows_x86_64
Hi there. These WUs are still
)
Hi there. These WUs are still being sent out by the servers. Please could someone on the project investigate.
It is not my computer at fault.
Here is the track of a typical wu. I also have many that have not been resent so they will possibly be resent to other people to waste their time and effort as well.
Workunit 384802385
Max # of error
)
Max # of error tasks: 20
That's one heck of a (wasteful) configuration.