I run Asteroids GPU (Nvidia RTX 2080Ti and 2070 Super) on a few machines, and found that the stock CPU setting of "0.01 CPU + 1 Nvidia GPU" is too low.

I think a CPU setting of 0.24 to 0.30 is much more accurate, in terms of fast CPU (3,75Ghz and up) and GPU (RTX or equivalent or higher); or slow CPU (2,50Ghz or below) and slow GPU (GTX or equivalent below this).

This will allow 1 core to be allocated per ~3-4 GPUs.

More testing is needed, but I believe the setting needs adjustment!

I stand corrected, some of the new WUs only utilize less than 5% of CPU.

It has been a while, so I thought I would compare the new CUDA version on different cards, and in particular compare it to the CPU version for efficiency.
Both of my machines are running Ubuntu 18.04.4 with the 440.59 (CUDA 10.2) driver.

Asteroids@home 102.15 Period Search Application (cuda102_linux)
GTX 1060: 72 watts, 33 minutes (0.55 hours) average per work unit => 40 watt-hours per work unit
GTX 1070: 105 watts, 24 minutes (0.4 hours) average per work unit => 42 watt-hours per work unit

Asteroids@home 102.13 Period Search Application (avx_linux)
Ryzen 2600 virtual core (all 12 cores running Asteroids): 7 watts, 1 hour 52 minutes (1.87 hours) average per work unit: => 13 watt-hours per work unit.

THEREFORE: The CPU work units are 3 times as efficient at the GPU work units.