Week ago, I had some work in the queue, but since nothing. Is the server OK or why is it like that ? No project, no work ?

Plenty of work to be had. https://www.gpugrid.net/server_status.php

Over 5500 tasks to be crunched still.

You are not asking for the available work obviously.

The only readily available work is the Python on GPU work.

You need to select that in your preferences.

More likely your other projects have preempted getting any GPUGrid work.

You also have a lot to task errors with illegal CUDA memory calls. You need to investigate your card.

What do you mean by illegal memory calls ?
Yes, I got a lot of errors and don't know why. Using Gaming driver, recently swapped for Studio driver and also installed CUDA 12.
But still error compute...

The application is calling out illegal memory calls. Look in the stderr.txt output of your errored tasks on the website.

I don't know anything about Windows. But Windows causes its own problems with these tasks.

I would say bad driver or overheating or overclocked card to blame.

oxley, you might try suspending all your other projects and see if you still get errors. I had similar errors appear when the latest round of World Community Grid WUs started being run on my hosts.

They also run fastest when they get to use >80% the BOINC allotted cpu time.

This means that when running other projects concurrently you should keep the manager options/ computing preferences.../usage limits/cpu usage around 20-40% so as not to overload the cpu, because Pythons will use ~60% of all available cpus no matter what the setting. They are not controlled by that BOINC manager setting so you can use it to keep the other tasks out of their way.