I read (while considering both for numerical routines) that the CLR could more effectively generate efficient instructions for arithemtically dense code and val

I read (while considering both for numerical routines) that the CLR could more effectively generate efficient instructions for arithemtically dense code and value types, including for generics, where the JVM uses simpler approach involving type erasure and treats built in types distinctly from user objects. JVM originated for Java which came earlier, so it’s not a fair comparison. And I’m not an expert here.. I stuck to C, CUDA and Oython in the end.