Benchmarks: Training DonkeyCar

Hello community,

I’m very interested in benchmarks for training DonkeyCar.
What kind of hardware are you using and how much images are processed per second ?

e.g.

Hardware,CPU/GPU,Images/s
RPI1,CPU,15
RPI2,CPU,?
RPI3,CPU,?
RPI4,CPU,?
Jetson Nano,GPU,???
MacBook Pro i5,CPU,?
Gaming Notebook, GPU, ?
AWS, GPU,?
GoogleCloud, GPU,?

Background: I try to find a cheap training environment for events or for our team. Many young guys have powerful gaming hardware. Maybe, we can find a way to use such resources in a common network cluster (mixed Linux, Mac, Windows).

Thanks for publishing your results

Cheers
Chrimo

BTW: I know, benchmarks need much more definitions.

What about to define and prepare an own dataset for DonkeyCar (from a typical race event) ?
This could be the reference measurement time on a standard RPI3.

Who has more experience in benchmarks and can give us an idea, howto compare systems running DonkeyCar or training components…

Do I need CUDA or is OpenCL or ROCm an option ?