C# and by far.
C++ has gcc and clang for compilers, as well as a few different IDEs, while C# has mono and the monodevelop IDE. Contrarily to some resilient myths, mono is not "unpolished" or "dangerous" or such. It's every bit as polished and alive as any C++ compiler on Linux, and a quality solution used by professionals.
there is MONO, but it's a port, it's not pure C#.
Mono is an implementation of open standards just like gcc or clang. I don't know what you mean by "pure C#", as Microsoft's .NET itself is written in C++.
In the process is being so, it is also far slower. Where as C++ runs code that is compiled directly for the CPU, C# runs as a managed language
C# is also compiled to machine code, it just does so usually "Just-in-time" rather than always ahead of time like C++. While C# does make compromises on performance where C++ makes none, it still offers comparable level of performance in most cases, and saying it's "far slower" is an exaggeration. I'd even say that it's easier to write fast code in C# than C++, leading to faster programs given the same time constraints.
Of course if what you're writing is scientific high-performance computing code, then C# isn't what you should be using, but I'm talking about the general case.
Another big advantage is that the whole library is developed under strict control in terms of runtime complexity - a feature you'll never be able to find under C♯
I've heard that from Herb Sutter and it's quite amusing. Yes, most of C++'s library has its complexities documented, but maybe because there isn't much more than containers and algorithms in the C++SL... Containers and algorithms in .NET also have their algorithmic complexities documented, but there is no point documenting the algorithmic complexity of a Printer or Socket class.
There are pointers in C#
But if you want to learn about pointers, you have to go with C++.
. And reference vs value types is basically the same idea as pointers vs values in C/C++.