I realize RAPIDS overall (especially model fitting, etc.) requires a certain level (or above) of NVIDIA GPU chips -- but curious what the cuDF-specific requirements are? Just via trial and error, I already noticed it seems to work on a broader set of chips than RAPIDS overall.
Is there a compatibility matrix available specifically for cuDF and which chips are supported?
We currently support CUDA 10 and newer and compute capability 6.0 and newer (Pascal+).
Seems this is not documented anywhere I can find. Let's change this into a doc bug.
Most helpful comment
We currently support CUDA 10 and newer and compute capability 6.0 and newer (Pascal+).