| Pemalite said: Ergo... Asynchronous Compute is exposed in Direct X 12. It doesn't matter how extensive that exposure is. |
There's a difference between 'exposed' and in the 'specification' in the eyes of specification ratifiers like Microsoft and what is defined so far is an 'interface' that can be used for stateless compute (async compute) ...
| Pemalite said: Of course Microsoft sits on top of the food chain. |
You're right, NEC did design a graphics chip and it was licensed by Intel in 1982 but that was a one time deal before DirectX (1995) was even created! (They might've held patents before but I imagine those are already expired since they didn't quite make it to the 21st century in the videocard market and Intel designed their own graphics chips shortly thereafter so I doubt NEC had a say in the DirectX specs.)
The main players (AMD, Intel, Nvidia) have already licensed S3TC so Microsoft doesn't need S3 graphics anymore and they won't be able to bully new players (Qualcomm, I'll be interested in knowing how they'll perform in 32-bit Windows games with high CPU emulation overhead) anymore either (such is the tragedy of being expendable) ...
The golden days of patenting computer graphics technology is over (I imagine we've still yet to see peak for other fields such as biotechnology) ... (the last big patented computer graphics technology appears to be hardware implementation for conservative rasterization and Volta's independent thread scheduling)
| Pemalite said: AMD can compete without it just fine. Hell. They have been. AMD's lack of a walled garden isn't the reason why they are in the current situation that they are in. It's because they have been rebadging Graphics hardware for over half a decade and not innovating. |
Not sure I see a problem with rebadging since that's mostly dependent on how frequent microachitecture designs are ... (AMD created 4 more graphics architectures since GCN and for Nvidia 4 more graphics achitectures since Kepler so AMD is pushing out new graphics architectures just as often as Nvidia is. Moreover it looks like AMD has stopped rebadging this round since their new baseline appears to be Vega all around with Vega 10, 11, 12 and Raven Ridge.)
I'd very much point the lack of a walled garden as a reason why AMD isn't performing when Vega is like the R600 where unused features that keep piling up takes up a significant portion of the silicon and power budget ... (Heck even the first GCN microachitecture isn't getting all it's features used in most of today's AAA games and I think that's the error AMD makes by not designing around current software.)
AMD needs Eric Demers (he works at Qualcomm) back ... (Raja Koduri isn't going to be their Jim Keller of GPUs)
| Pemalite said: Even on PC. FP16 is a rarity on the hardware side, which you have alluded to. Untill it reaches mass adoption across the industry, it's support will be relatively niche'. |
It's less of a rarity than you think, practically all current AMD (GCN3+) and Intel graphics (Gen 8+) support FP16 in addition to PS4 Pro ...
Even if the implementation is not double rate, FP16 can still benefit from register savings thus increasing occupancy and performance ...
You seem to think that the industry is centered around Nvidia but Nvidia's only big when you look at Windows PCs however that's not so for other vendors (Apple, Sony, Xbox) or integrated graphics (where Nvidia has 0% marketshare over there) all of which takes devs a considerable portion of their attention ... 
AMD only needs to target FP16 optimizations for AAA games to fight their hardest battles but all the merrier if there are other parties (Apple, Sony and Intel) who are in favour too ...
