AMD like NVIDIA? Dedicated hardware on future Radeons to handle upscaling?
As reported by the Twitter user Kepler_L2 and reported by Videocardz, AMD updated the LLVM repository on GitHub for better support for future architecture-based video cards RDNA 3also known internally as GFX11. Among the novelties are the instructions WMMA (Wave Matrix Multi-Accumulate), used for data processing by the algorithms of machine learning.
The WMMA instructions would seem to imply the presence in the RDNA 3 architecture of hardware units dedicated to upscaling management, like the Tensor Cores of NVIDIA’s GeForce RTX 2000 and 3000 GPUs. At the moment, however, there is no other information, and therefore everything must be taken with due precautions.
After the major improvement of the FSR with the move to version 2.0, AMD seems to have even more ambitious goals with the FSR 3.0. However, a question arises: how will this choice affect developers’ use of FSR? One of the aspects that marked the success of FSR concerns above all the compatibilityas it does not require dedicated hardware (unlike NVIDIA’s DLSS) and being open sourceallowed for much simpler as well as greater integration accessibility for users.
Not by chance FSR adoption was twice as fast compared to that of the DLSS. Should AMD also transition to an implementation tied to dedicated hardware units, the underlined benefits may vanish.
We will see, it cannot be excluded that FSR continues to remain agnostic, it should not be forgotten that Intel’s XeSS promises to be able to exploit the XMX units of the Alchemist GPUs and, through a specific instruction, also work on competing GPUs.