| Soundwave said: DLSS means the GPU has to be either Turing or Ampere-based because older Nvidia generation GPUs (such as the current Switch Mariko chip or even the Pascal chip) can't perform DLSS. |
Not true.
You can do the DLSS calculations on the CUDA cores.
Keep in mind that Tensor cores are just multiplying FP16 matrices, then adding the multiplication to the fp32 matrix and then to the accumulator.
The Tensor cores can also do INT8 and INT4. (Integers) which can also be performed on the CUDA cores.
To put it simply though, think of CUDA cores as slower, more precise, more flexible cores, but the Tensor cores as very specialized, no-bullshit, simple, less precise, but quick cores.
They can both technically do the task of the other, but generally you wouldn't want to due to performance.
Basically if Nintendo wanted... They could take the Pascal based Tegra, shrink it down on say... Samsungs less congested 8nm process and clock it extremely high to make up for the compute overhead.
| Peh said: Please no OLED. Looking at the flagship smartphones by Samsung I've got, they all ended with burn-in issues. IMO, I would prefer LCD. |
LCD is definitely preferred for longevity.
But OLED has vastly superior quality.
How are you getting burn in? Always-on display?

www.youtube.com/@Pemalite








