Forum > Windows

Computation Error with CUDA

<< < (4/4)

Raistmer:
It will show that on web page by adding corresponding message to stderr.
You can edit/install <flops> field for this app_version.

Frizz:

--- Quote from: Raistmer on 10 Aug 2010, 07:39:56 pm ---It will show that on web page by adding corresponding message to stderr.

--- End quote ---

Unfortunately SETI is down for the next - at least *g* - 3 days.

But still it would be nice if the BOINC client could show this.


--- Quote from: Raistmer ---You can edit/install <flops> field for this app_version.

--- End quote ---

I guess FLOPS causes the -177. After I removed FLOPS its working again. Ah those estimates ... a pain in the ass!

Richard Haselgrove:

--- Quote from: Frizz23 on 10 Aug 2010, 07:35:26 pm ---
... It would be nice if the BOINC client would show this - and not just "Computation error". Especially at times when SETI is offline.


--- End quote ---

Request reported on boinc_alpha. (I agree)

Josef W. Segur:

--- Quote from: Frizz23 on 10 Aug 2010, 07:43:13 pm ---I guess FLOPS causes the -177. After I removed FLOPS its working again. Ah those estimates ... a pain in the ass!
--- End quote ---

Yes, the rsc_fpops_bound divided by flops is the amount of time BOINC will let the task run before killing it, the bound is 10 times the estimate which under normal circumstances is more than adequate.

An alternative to removing flops would be to use Fred M's new rescheduler. It can push all the bound values up and thereby cure the -177 errors. I don't remember the details exactly, but if you check "Limit rsc_fpops_bound" on the Expert tab, then click the Test button, it will fix those bounds without actually rescheduling any tasks. What I'm not sure about is whether there's another step like maybe manually copying the client_state.xml produced by the test back into the BOINC data directory or such.
                                                                                Joe

Navigation

[0] Message Index

[*] Previous page