Some details at last.
They finally updated the rasterizer and dispatcher:
https://www.techpowerup.com/reviews/AMD/Vega_Microarchitecture_Technical_Overview/images/arch-18.jpg
https://www.techpowerup.com/reviews/AMD/Vega_Microarchitecture_Technical_Overview/images/arch-19.jpg...
Most of the major titles are both Xbox and Playstation.
It really comes down to something being an AMD or Nvidia based feature and how they push(or don't) adoption. Nvidia is much more aggressive and consistent, while AMD is mostly passive with short campaigns but lacking persistence.
Waiting...
When you consider GCN designs usually have 4 ROPs per 32b memory channel, a 96/128 ROPs design has some fun implications: 32 ROPs - 256b(8 channels), 64 - 512b(16), 96 - 768b(24), 128 - 1024b(32).
There would be slight issues with power and board cost.
Memory reads and writes.
Empty connector pads and all those test points.
It's a pre-production card you'd seed to developers/companies. Explains the pro/game driver switching. Doesn't explain why ship it retail. Not something you'd rely on for production or gaming.
Thats the exciting question.
There is no "on paper" at the moment. AMD haven't released anything showing layout or implementation details. And with geometry binning seemingly off by default, we can't ferret out any information on it.
Lets use Polaris since its the most recent known quantity. The theoretical triangle...
For the most part it was at the defaults, except floats per vertex(slider a bit over halfway). However, there was some rendering initially with everything at default.
I was one of the people in chat asking for trianglebin and it didn't seem to be doing any tiling.
Standard GCN behavior, fully render one triangle, render next, etc.
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.