A bug of NVCC compiler

Accelerated Computing CUDA CUDA NVCC Compiler
1

I write a simple C++14 program which can be correctly compiled by gcc, but cannot be compiled by nvcc. The minimal reproducible code is shown below:

#include<cstdio>

template <int K>
struct Struct {
    int a;
};

template <int K, int... Ks, typename F>
void apply(F func, int KK) {
    if (KK == K) {
        Struct<K> i{K};
        func(i);
        return;
    }
    apply<Ks...>(func, KK);
}

template <typename F>
void apply(F func, int KK) {}

int main() {
    auto fun = [](auto i) { printf("enter %d\n", i.a); };
    apply<1, 2, 3>(fun, 1);
}

The compilation error is shown below:

error: no instance of overloaded function "apply" matches the argument list
            argument types are: (lambda [](auto)->auto, int)
          detected during:
            instantiation of "void apply<K,Ks...,F>(F, int) [with K=3, Ks=<>, F=lambda [](auto)->auto]"
(15): here
            instantiation of "void apply<K,Ks...,F>(F, int) [with K=2, Ks=<3>, F=lambda [](auto)->auto]"
(15): here
            instantiation of "void apply<K,Ks...,F>(F, int) [with K=1, Ks=<2, 3>, F=lambda [](auto)->auto]"
(23): here