Page 1 of 1

inconsistency device numbering setup/logging

Posted: Thu Jun 09, 2016 1:19 pm
by boris
jim can you please tell me the logic in gpu device numbering?
I try to locate the unstable gpu and have some troubles..
I am on version OctaneRender_for_3ds_Max_2.24.2_-_2.13
- in octane devices setup my gpus are numbered 1 to 6.
- 1 is dedicated and disabled.
- logging reported chain reaction fails starting with device 1 following 2,3,4 and 5 unordered

does the logging renumber only enabled cards? and are devices basically sorted by bus ID number?
I need to get my sheet completed with octane's numbers:
Bus ID 04 - gpu shark: 1 - HWinfo: 0 - memtestG80: 0
Bus ID 05 - gpu shark: 2 - HWinfo: 1 - memtestG80: 1
Bus ID 06 - gpu shark: 3 - HWinfo: 2 - memtestG80: 2
Bus ID 07 - gpu shark: 4 - HWinfo: 3 - memtestG80: 4
Bus ID 10 - gpu shark: 5 - HWinfo: 4 - memtestG80: 5
Bus ID 11 - gpu shark: 6 - HWinfo: 5 - memtestG80: 3

trying to trigger the render failed with isolated card rendering while memtesting the other 4 to get the surrounding temperature...
stress_test.jpg
Sorry I know you got probably better to to but can you shortly clear up how the device numbering works or tell me why the numbers in the device setup are not the same as in the octane log?
tnx

Re: inconsistency device numbering setup/logging

Posted: Thu Jun 09, 2016 8:29 pm
by mojave
As an option you could use the standalone version and disable all GPUs but one (go to File > Preferences > Devices) until you find the one causing the issue.

Re: inconsistency device numbering setup/logging

Posted: Fri Jun 10, 2016 7:57 am
by boris
mojave wrote:As an option you could use the standalone version and disable all GPUs but one (go to File > Preferences > Devices) until you find the one causing the issue.
tnx mojave but this is what I basically was doing: letting octane render with ONE gpu while stress testing the other four (to simulate the heat that would occur when rendering with all gpus.
anyway the memtest reported some errors writing random blocks on two of the surrounding cards.
NOW two weeks ago I installed NVIDIA driver 365.19 for being able to test octane 3.
after reverting the driver back to 347.25 the problems seem have to be disappeared: just reached 12500 passes at 5200x3200 resolution after 15 hours of rendering with ALL gpus WITHOUT render failed warning.
so beware of that driver..

What I basically wanted to say: I makes NO SENSE to me renumbering the rendering gpus for the octane log. IMO they have to have the same ID in the log like in devices setup. Otherwise tracking down issues like i ran into can be confusing.
and yeah, time stamps in the log also would be helpful :) not sure maybe they are already implemented in 3.0 version by now.