On the title, FAIU, the techs have set a very high fault allowance, meaning devices are permitted to have many errors **, so it's a consequence that more copies need to be circulated before a quorum is reached.

The time [FLOPS] is build on last few days historical data, so it is in principle pretty accurate from that perspective.

The WCG server has send an instruction to clients to not use the DCF, so projections should in principle be pretty stable for Ready to Start work.

Question: Is this 1:46 a "run it exclusively by itself", or when "run with many concurrent", a setting of e.g. 12 concurrent on one GPU? Also, do you run a mix of CPU and GPU versions of HCC? Finally, what client version exactly?

edit: ** To clarify, high fault allowance is like 80%+ [or was it 85-90%?], and cards still not being blacklisted, which is understandable in a way given that a successful task runs 18 times faster [if having the GPU to itself]. Do not know though for sure if that's any part of the motivation to be highly tolerant of low success-rate devices.