我在國外論壇上所發表的評論內容如下,我只簡單用中文講一下重點：DLSS3肯定還是得先降低資料處理量(也就是降低解析度)，才可能實現加速，因此DLSS3仍然還是一種放大技術。DLSS3所用到的AI技術主要是針對運動的預測,但只要使用者突然改變運動方向(很常見吧?),就會出問題
To be more correct, in addition to zooming in, DLSS3 also uses Optical Multi Frame Generation to generate entire individual images based on the frame before zooming in. Optical Multi Frame Generation still operates at reduced resolution. Without first reducing the amount of data that needs to be processed, which reduces the resolution, it is impossible to achieve acceleration. In other words, the Zoom technology of DLSS3 is better, but the image quality defects brought by DLSS2 in the past will still exist, but the degree of defects in DLSS3 will decrease.
The AI technology of DLSS3 is mainly based on the analysis and prediction of the motion of all objects in the game, but as long as there is an unexpected operation behavior, the image defects caused by DLSS3 will appear. That's why the DLSS3 images shown by nVidia are all straight forward, because the inertial operation of straight forward is unlikely to predict wrong. However, after maintaining a straight line for a period of time and making a sudden turn, the AI will make mistakes, but it is normal for the operator to change direction.
nVidia霸主地位快要不保了
https://www.dcard.tw/f/pc_game/p/240032365
在這篇文章中,已經很明確地提到,AMD RDNA3是首款採用MCM GPU的顯卡晶片,他是利用真實存在的chiplet小晶片,並利用Chiplet小晶片的低成本特性，透過某種程度上可稱為『堆料』的手段，來實現真實的加速運算,提高真實地提高幀率Frame rate
相對地,nVidia完全沒在未來大趨勢的Chiplet上做投入,只剩下用插幀Frame技術來誤導消費者....nVidia霸主地位真的要不保了
