fatslob-:O said:
Like I said, "interconnect bandwidth means jack shit". What part do you not understand ? |
bandwidth is important more than you think, if not why would AMD and nvidia increase the internal bandwidth of the gpu generation?
is imprtant for many things like vertext texture fetches and other things, and i dot say this, its expert people on the topic, like people from nvidia
but if you at least read the whole article from gaming blend you will see that he doesnt say this himeself but that consulted people for this
want their comments?
here
http://developer.amd.com/wordpress/media/2012/10/Tatarchuk-Tessellation(EG2007).pdf
"
Meshes are Expensive Beasts!
•Want to express more and more details
•
Complex, rich worlds of recent games require detailed characters
•
Shading has evolved tremendously in the recent years
•Meshes are a pricey representation
•
Need to store a lot of data (positions,
uv
s, animation)
•Vertex transform cost is incurred
regardless of how close the object is
to the viewer
•
Expensive for complex shaders or animated / detailed objects
•Novel GPUs’ unified shader archit
ecture has much more efficient
geometric processing
•
But memory storage and fetch bandwidth is still a big concern
•
Especially for animation
•
Large meshes cause less vertex cache reuse
"
We would like to change that now!
Both must manage details for stable performance
Bring tessellation techniques
from film in real time
rendering scenarios
• Takes advantage of highly efficient tessellation HW and
memory bandwidth of ATI Radeon HD 2000 Series
• Fast displacement mapping and animation
"
or do you prefer nvidia?