There are a few open source games that can tax a more advanced GPU than the Pandora has at least; games like OOlite and various 3D engines that run commerical game wads. But yeah, other than those exceptions, most things we'll be running on our Pyras are going to be CPU bound, not GPU.
Things are CPU bound until they aren't. At some point the CPU ceases to be the bottleneck.
Example: If your use case is surfing the internet, any CPU of Core 2 Duo or later vintage will do just fine. Processors prior to that cross an invisible boundary below which things start getting bad in a big hurry.
If we're going to seriously discuss a successor Pyra SoC, we're entering an age where the CPU side of the SoC dies are, if anything, way overpowered (mostly in core count). On the Pyra we will generally need 1 relatively fast core for the foreground application and maybe half of a second core to take care of background tasks. Modern ARM SoCs are tending towards 6-10 cores of varying speeds. Unless you can point to truly multi-threaded applications that will run...?
So, that one primary main CPU core will likely have some sort of cap on it. I realize that it isn't this simple, but people understand Ghz even if they don't. So, if two SoCs both have more or less equivalent magnitude primary CPU cores (1.7 to 2.5 Ghz) but one has a substantially "better" GPU (speed, driver accessibility, etc.), the exact speed of the CPU portion may then become be less relevant.
An industrial ARM SoC might have a pile of blazing fast cores (and the thermal footprint to match) but near enough to nothing in the GPU department.
A mobile ARM SoC will likely be more balanced. It gets specs credentials for the CPU core count and Ghz of it's fastest core, but needs the GPU to be good enough to feel speedy. 99% of the time the 3rd through Nth CPU core will be idle.
A modern automotive SoC will have several CPU cores, but much more emphasis on the GPU. It will be expected to drive fluid graphics on upwards 6+ displays simultaneously.
Example using R-car H3:
[doublepost=1529438772,1529437199][/doublepost]As noted above, processor efficiency (battery & thermal impacts) are generally better the smaller the lithographic process is.
The OMAP5432 is built on a 28nm process
The RK3399 is built on a 28nm process
The R-car H3 is built on a 16nm process
However, the A72 in the RK3399, if it were made on the same process, would be faster core to core at the same Ghz than the A53 that the R-car H3 uses as it's primary core. But, the above lithography difference in efficiency might well make up for that - I don't know.
Which brings me to the question of what is out there with a <= 16nm process and an A72 or better core?
Can we get a Helio X30?
10nm process...
mobile so balanced CPU/GPU
graphics?
Planet Computers was able to get the X27 for their Gemini - and it is running both Android and Debian Linux, so there has to be some taste for niche products and Linux.