Don't underestimate how competitive GPU market once was. That TBDR concept, I believe, is not something new coming out recently. [...] If one concept is truly superior to others, industries usually adapt to that.
TBDR is certainly not new, but it's more complicated and therefore more challenging to implement. Historically, there was a split where mobile went with TBDR (because of it's superior energy efficiency) while desktop went with forward rendering (because it was easier to implement and scale, while energy efficiency was a secondary concern). TBDR traditionally suffered from front-end bottleneck (geometry thoughtput), so it was optimal for mobile games which featured lower-quality graphics anyway.
As to industries adapting... if you have literally decades of R&D invested into a particular approach and it works very well, not much point in pursuing an alternative with unclear benefits. Apple is in a slightly different position, since they currently have fastest TBDR GPUs on the market — it makes perfect sense for them to build on that technology.
In the end, TBDR comes from the need to do "more with less" — it was developed to overcome the memory bandwidth limitations on slower, power-constrained hardware. This is less important in a desktop (or even laptop) situation. And TBDR is certainly not a magic bullet. Yes, it allows one to get better performance with slower memory. But this will only apply to games. It won't do anything for compute tasks for example. What is interesting about Apple's approach is not that they are leveraging TBDR. It is that they are opening the architectural details to the developers. You have direct access to the on-chip tile caches, which allows one to code some algorithms much more efficiently than it would ever be possible with forward rendering.
The level of performance shown by RTX 3080 and 3090 seems staggering.
And so is the power consumption. Some early benchmarks suggest that the 3080 might be somewhere between 60-80% faster than the 2080... but it also consumes 50% more power! If you normalize for power usage, you get about 10-20% improvement... which is still great but suddenly not as impressive looking to an average gamer. Which really tells you about what sells at the market. Current Ampere designs are deliberately designed to be power-hungry monsters simply because people will buy that.
I think it's a bit sad that the outstanding engineering achievements are banalized by the fact that the biggest gains are still because of fatter, more power-hungry chips.
P.S. By the way, one particular thing where I am grateful to Apple is that they didn't join the other with all the "shader core" nonsense. They just say — "our GPUs have 8 cores and that's it" instead of "512 cores" like everyone else does.
