Oh dear, AMDs DLSS equivalent, may not work on the 6 series.
Go to 2:11.
This is great, now we have Machine Learning Accelerators, even better than tensor cores.
What these things are doing is to run some matrix maths called tensor. And they are doing very fast but they are not the only way to run this type of maths. It can even be done on the CPU (at a much lower speed). Or on the shader cores in the GPU.
Unlike a GPU chip, adding dedicated hardware to a chiplet is a logical step. But unless Direct ML becomes a standard, AMD can put thousands of Machine Learning Accelerators on their boards and they will be useless. With Direct ML such hardware can be used for faster upscaling and denoising.
Gamer Meld is stupid, the 6000 series can also do upscaling but it won't do it as fast without "Machine Learning Accelerators".
https://www.techspot.com/article/2151-nvidia-ampere-vs-amd-rdna2/
"As previously mentioned, the Compute Units in RDNA 2 now support more data types; the most notable inclusions are the
low precision data types such as INT4 and INT8. These are used for tensor operations in machine learning algorithms and while AMD have a separate architecture (CDNA) for AI and data centers,
this update is for use with DirectML.
This API is a recent addition to
Microsoft's DirectX 12 family and the
combination of hardware and software will provide better acceleration for denoising in ray tracing and temporal upscaling algorithms. In the case of the latter, Nvidia have their own, of course, called DLSS. Their system uses the Tensor Cores in the SM to perform part of the calculations, but given that a similar process can be constructed via DirectML, it might seem that these units are somewhat redundant. However, in both Turing and Ampere, the Tensor Cores also handle all math operations involving FP16 data formats.
With RDNA 2, such calculations are done using the shader units"
.