NVIDIA RTX series graphics introduced new architectures with elements designed to accelerate different types of operations. In addition to the CUDA Cores in charge of “pure” graphics acceleration, they had the RT Cores in charge of certain calculations for Raytracing optimization, and the tensor cores specialized in processing the inference of Deep Learning systems (for example, DLSS).
Although AMD already introduced hardware support for raytracing in RNDA 2, RDNA 3 (codenamed GFX11) will add a technology similar to NVIDIA Tensor Cores named WMMA(Wave Matrix Multipluy-Accumulate), capable of performing calculations on 16x16x16 matrices.
This architecture will have an API in the C++ programming language to be able to divide the problems of this type of matrix into blocks and distribute them throughout said cores, processed in parallel.
This could leave room for AMD to add some AI acceleration to its FSR technology in a hypothetical FSR 3.0 release, although FRS 2.0 already achieves amazing results without it. However, this technology could be applied to other calculation accelerations related or not to the inference of Deep Learning networks.
End of Article. Tell us something in the Comments!
Antonio Delgado
Computer Engineer by training, writer and hardware analyst at Geeknetic since 2011. I love to gut everything that passes through my hands, especially the latest hardware that we receive here to review. In my free time I mess around with 3d printers, drones and other junk. For anything here I am.
Add Comment