what I noticed this afternoon: tasks from the OPNG_0133848 series get some more credit points: +/- 900.
Processing time also is a little bit longer.

The ideal way for credit to work is to formulate an proper and accurate estimated flops value for each work unit. Then assign credit based on runtime and the flops performance of the device. This way more difficult tasks (more flops = more computation time) will automatically be awarded more credit. And it should scale appropriately if the device flops are properly recorded client side.

Or

Standardize the WU size, so that ALL WUs are the same size and difficulty computationally, then assign static credit reward. That way no matter what device completes it, it gets the same credit, and faster devices earn more points relative to how many tasks they complete. This is a very fair system and will remove the cheating element by client side flops reporting manipulation.

Both options require some adjustment from the project’s side in how they form their WUs.

did you change anything? new driver install? could be a lot of things really.

no, I didn't.

But no further "Error" cases since yesterday afternoon.
So let's keep the fingers crossed :-)

Edit: one of my PCs got new WUs just a few minutes ago :-)