Parallel Array Architecture for a Graphics Processor

專利原文。

看起來主要著重於content switch overhead、dynamic warp formation的cost下降，可以更有效率地同時執行CUDA、PhysX和graphic功能等。

基本上可以說是NV50/G80~GT200目前各種缺點的改善、架構上的refine等等，來對應需求規模更大的DX11。

不過反過來說，並沒有看到MIMD之類的結構，也就是說嚴格說來改動沒有想像中的大。

http://farrarfocus.blogspot.com/2009/04/hierarchical-processor-array.html

Hierarchical Processor Array

Timothy Farrar 的一些意見。頗具參考價值。

—-

http://developer.nvidia.com/object/bindless_graphics.html

Bindless Graphics Tutorial

http://farrarfocus.blogspot.com/2009/04/nvidias-bindless-graphics.html

NVidia’s Bindless Graphics

Two awesome NV GL extensions which can enable a 7x draw call speed up by removing the GL bind pattern which causes the drivers to be cache miss bound. Until ARB gets around to API fixes that remove this problem for good, it is great that NVidia is supporting a work around!

—-

http://farrarfocus.blogspot.com/2009/04/irony-in-ps3-vs-larrabee.html

Irony in PS3 vs Larrabee

http://farrarfocus.blogspot.com/2009/04/caustic-graphics.html

Caustic Graphics

More info has surfaced on the Caustic Graphics Hardware.

http://hpc.pnl.gov/people/fabrizio/papers/ipdps07-graphs.pdf

Challenges in Mapping Graph Exploration Algorithm on Advanced Multi-core Processors