|
主要是因为GPU虽然有部分流处理器的特征,但是很多数据还是有机会重用的。比如(Aniso) Texture filtering, large kernel filtering之类的。使用cache, 可以减少带宽的浪费。至于延迟,倒不是非常的敏感。
但是GPU ...
RacingPHT 发表于 2009-6-16 10:12 ![]()
Not sure what yiu meant here. If it is a graphic app, multi-level caches provide sharing data among pixels. If a computing app, shared memory is doing the sharing job for threads within a block. |
|