It's been possible for quite some time.

In the early 2000's there was a book on using Direct3D from C# that was pretty influential as far as changing people's assumption that you couldn't do high performance graphics in a GC'd language. In the end a lot of the ideas overlap with what c/c++ gamedevs do, like structuring everything around fixed sized tables allocated at load time and then minimal dynamic memory usage within the frame loop. The same concepts can apply at the graphics API level. Minimize any dynamic language overhead by dispatching work in batches that reference preallocated buffers. That gets the language runtime largely out of the way.