Maximum number of emitted rays per second

I don’t think the marketing material contains that maximum rays/second number.
It mainly depends on the number and generation of RT cores, cache sizes, and the memory bandwidth.
Overall ray tracing performance inside applications then also depends on the generation, number and speed of the streaming multiprocessors (CUDA cores) and memory accesses which is affected by the cache sizes and memory bandwidth again.

The specification or datasheets of the individual GPUs or the architecture documents of GPU generations contain TFLOPS/second numbers for the different core types which can be compared.

Please have a look at Compare NVIDIA RTX Graphics Solutions section here for current Ada, Ampere and Turing workstation boards for example:

In general, the newer and more RT cores, the higher the memory bandwidth (the wider the memory bus), the bigger the caches, the more streaming multiprocessors, the higher the clocks, the more VRAM, the better is the board for GPU raytracing.

Here are links to tables with most of these numbers for the different boards:

A very coarse rule of thumb is that each RTX GPU generation doubled the ray tracing performance.
That is not always true for all ray tracing features, for example, motion blur on triangle data improved 5-fold in Ampere over Turing in actual applications (and even more in dedicated tests) and curve primitive intersection improved similarly.

So if you’re looking for the highest ray tracing performance, pick the Ada GPU based board with the biggest product number you can afford. Today the RTX 6000 Ada Generation board is the highest spec. RTX 4090 when you can live with consumer grade products.