I wouldn't worry about overall credit as long as it is fixed going forward as you mentioned.

You're one of the most communicative and responsive Admins of Projects around

That does sound like a problem. I do know that the WUs for the new batch should take longer, so the credits should be higher. I might have inverted the scale factor.

I am currently on vacation, so I won't be able to fix this until I return in 2 days.

So my 4070 ti, which takes 3, 4 minutes per task, runs about the same number of tasks per hour?

RTX 4080 Super  10min 17sec, 3 at a time = 420 per day.
RTX 4070TI Super 7min 56sec, 2 at a time = 363 per day.
RTX 2060 Super  11min 26sec, 1 at a time = 126 per day.

If I buy a high-performance FP64 graphics card, such as an Nvidia Tesla V100 or P100, will there be a big difference in performance?

Can someone explain why my work at NumberFields@Home is not showing up in boincstats (https://www.boincstats.com/)?

Mean rainfall (mm)
The arithmetically averaged total amount of precipitation recorded during a calendar month or year. Both mean and median (decile 5) rainfall are included in these statistics, although from the meteorological point of view the median is usually the preferred measure of 'average' or 'typical' rainfall. This is because of the high variability of daily rainfall - one extreme rainfall event (such as a slowly moving, severe thunderstorm) will have less affect on the median than it will have on the arithmetic mean.

I think I figured out the root cause of what you are seeing. I am basing the credit on the ratio of average CPU times from one batch to the next. So the RAC for just the CPU will be almost dead on. However, it appears the efficiency of the GPU app went down a little bit for this latest batch compared to the previous batch.

Going forward, after the current batch, I will use the better of the two ratios to determine credit. That way, if the RAC changes it will increase instead of decrease. It would be best to have different credits for CPU and GPU, but I can't do that since the same WUs are interchangeable between the two platforms and hence a single credit is associated with the WU.

The current case has a larger standard deviation than usual, so you will see greater fluctuations in your run times and RAC, but it should not be a huge difference. The remaining time on an unstarted WU in the manager will give you an idea of what the average runtime is.

I'm not sure what they were testing in those graphs, but every app will be different.

With my AMD Ryzen9 7950X: my CPU temps were about 90C with HT on, and about 80C with it off. I don't like stressing my CPU that much so I leave it off.

One of the compute driver bugs with the Intel Compute Runtime on Battlemage appears to be around having much higher latency than Alchemist and the NVIDIA/AMD competition. Here's a look at the OpenCL kernel latency being many times higher on the Arc B580 than the Alchemist A-Series. Hopefully this Intel Compute Runtime issue can be quickly worked out.

More importantly, the error is that it can't "set the device", so it's probably some kind of environment problem, maybe a permission or SElinux thing.

So there are only 2 linux hosts with a 5000 series card (database only keeps last 5 days of results). The one mentioned earlier with successful results and this one with failures:
https://numberfields.asu.edu/NumberFields/results.php?hostid=2900384

process exited with code 1 (0x1, -255)

Setting GPU device number 0.
Cuda Error: failed to set the device.
Error: Failed to initialize Cuda.