[calaphos]

The hardware is heavily optimized for low precision matrix math, pretty much only used for AI.

[touisteur]

Not since the Ozaki scheme has appeared. Good high-precision perf from low-precision tensor units has unlocked some very interesting uses of low-fp64-perf GPUs.

[bobsmooth]

And in graphics rasterization.

[namibj]

Those parts (the tensor cores) aren't used for rasterization.

[david-gpu]

While raster units are separate from tensor cores, both can be leveraged for image rendering. The simplest example of this is Nvidia's DLSS.