>>106950581
They do complex matric operations, the issue is that currently devs dont have access a way to use them so you only see them when you use upscaling, path tracing, RT and frame gen but after Shader Model 6.9 is released for the next Direct X12 release on Q1 2026 devs will be able to start utilizing tensor cores and cuda cores directly from the SDK utilizing Cooperative Vectors which is why AMD is finally moving on to use them