For a long time, consumer Nvidia Graphic cards supportted having plain cuda cores despite not being branded for accelerators.
The problem in my case is while Tensor cores allow to perform cuda, they are restricted to matrix and less precise computations. This speeds up machine training and what depends on it like nvidia dlss.
On the new rtx series, I m no longer seeing any reference to the number of cuda cores (like rtx 3090 or rtx 3070).
Does it means things like deterministic cuda based rendering with Indigo Render now require Tesla gpu for newer cards?