I received some work units (tasks) today but most of them failed with a "error while computing" error.

The error occurs usually within 1-2 minutes and one task went 7-8 minutes. Another task is still running (80% done),

Error encountered on two Windows 11 PCs both running BOINC v 8.2.8 with different NVidia GPUs.

The first is an older PC where a task is still running: https://gpugrid.net/gpugrid/show_host_detail.php?hostid=648899 with 3 tasks failed and 1 still processing (type is ATM: Free energy calculations of protein...)

The other PC is a newer PC: https://gpugrid.net/gpugrid/show_host_detail.php?hostid=638123 with 3 tasks failed.

Having same issue.

---

It usually takes the researchers a few small batches of tasks to sort out the proper configuration parameters. Only when the small batches are mostly successful do they give a larger, longer lasting batch.

Thank you Keith! Glad to see progress and glad to help.

The first task did complete successfully. A second task last night failed after multiple hours.

Slot 0 was already occupied by another program, so it was aborted after 7 minutes.
Werkeenheid 31550481

WARNING: The script pyaml.exe is installed in 'C:\ProgramData\BOINC\slots\0\Scripts' which is not on PATH

Yes, the tasks are much harder to setup correctly on Windows. OTOH, in Linux it is much easier to get the configuration right because of the way that the OS sets up support files and applications. Still doesn't help you when the researcher plainly forgets to include a needed research in the task package and then attempts to use it when it isn't there. At least most of those fail fast, less than a minute. The irksome ones are the ones that seem to be running correctly but eventually hits a NaN error after many hours. Again, expect to have many errors for the first batch of work that gets sent out.

...The irksome ones are the ones that seem to be running correctly but eventually hits a NaN error after many hours...

I have had exactly those quite frequently within the past 2 days :-(

https://gpugrid.net/gpugrid/results.php?userid=125700&offset=40&show_names=0&state=0&appid=

It's a shame that all the p38_A15_A13 etc. don't even reach 1%, 17 succeeded, 22 failed, too bad it was always going smoothly.

It's a shame that all the p38_A15_A13 etc. don't even reach 1%, 17 succeeded, 22 failed, too bad it was always going smoothly.

as I already wrote in the GPUGRID Discord channel: I am surprised that obviously, before issuing a new batch, none of the tasks are being tested before being sent out.