Hey,
dont know if that should be like that but all "1.20 Find seeds with zero villages within a radius v1.08 (cuda) windows_x86_64" are failing on my Quadro M5000.
On my P2200 they are slow but working.

<core_client_version>8.0.2</core_client_version>
<![CDATA[
<message>
{Segment laden}
Eine VDM l�dt, l�scht oder bewegt ein MS-DOS- oder Win16-Segmentbild.
Eine Ausnahme wird veranlasst, damit ein Debugger innerhalb dieses 16-Bit-Segments Symbole und Haltepunkte laden, l�schen oder nachverfolgen kann.
 (0x2be) - exit code 702 (0x2be)</message>
<stderr_txt>
boinc gpu 0 gpuindex: 0 
No checkpoint to load
Maxwell device detected - running smaller kernel runs for compat.
GPUassert: the launch timed out and was terminated (code 702) main.cu 4890
</stderr_txt>
]]>

Link to one Workunit https://minecraftathome.com/minecrafthome/result.php?resultid=10540427

I believe the card is too old an architecture to run on the app. Check with boysanic the developer.

The m5000 is maxwell ( GTX 900 series equivalent i believe) I'd think it would run

Yes would think that too.
GTX 900 Series is not newest but also not extremely old^^
But got in 2 hours about 370 failed "1.20 Find seeds with zero villages within a radius v1.08 (cuda) windows_x86_64" works so it looks like actual it is not working.
Just wanted to if that is normal or a bug

The error message says it detects specifically a case for the Maxwell class gpu and makes some changes in the app code for it.
But it does not handle the card correctly after that and produces instant errors.
You still need to check with the developer boysanic and see if the app really CAN handle the Maxwell card and if it supposedly can, need to report to him the app is busted on your host.

Hi,

The message about "Maxwell detected" was specifically something I was trying to optimize kernel runtimes on older (and weaker) cards.
Not overall task runtime, but specifically shortening the per-GPU-execution time (of which there are many in a single task) as to try to resolve the "launch timeout" error.

I don't think we ever fully resolved it. But, I would expect that a M5000 could run it.

It's possible I need to do more work to get it running on GPUs that can't cope with the launch bounds we've given so far. I don't have access to a GPU in Maxwell besides a Titan X, but my Titan X runs these tasks without complaint.

I'll add this to my list to investigate further. Lately I've been focused on our CPU-based app, Xoroshigo, but it's running fairly stable after the fixes we deployed yesterday.

Oh, since this is a windows machine, something you can try to work around this is to change this registry value if you're comfortable with that.

HKEY_LOCAL_MACHINE\System\CurrentControlSet\Control\GraphicsDrivers
Add a "DWORD" key called "TdrLevel" , set it to 0.

Fair warning, if it's your display GPU, you may find your screen unresponsive if you run GPU-based tasks.
If this is a concern to you, you can set TdrLevel to something higher. I'd start with 3, then try 10 if that also fails.

Please let me know if you do try this and how it goes for you.

Yeah my screen output is on the igpu so no problem.
And this setting (with 0) seems to be working.
Workunit is only at 5%, but before it always was stopping with the error after a few seconds.
But i will post an update when it finished the task

Now it finished and everything is ok.
Wasnt really fast, but no errors.