statistical soundness: prover is computationally unbounded

arguments vs proofs: in arguments, as in SNARKs, prover is computationally bounded, specifically polynomial time while in proofs, prover is unbounded.

Another comparison in circuit satisfiability $C(x,w)=0$ problem is difference between soundness and knowledge soundness. where, soundness means, prover is trying to prove existence of a valid $X$ which satisfies the circuit whereas in knowledge soundness, prover knows $x$.
There can be cases where one make sense over the other. For example: soundness is needed when $V$ sends $C$ along with input to check whether input satisfies.
Knowledge soundness is needed when Prover has a secret $w$ along with computation, like proving knowledge of private key to a wallet.

Creating SNARK from Interactive proof:

simply sending the input $x$ to the verifier with circuit C to check the computation, not a SNARK as proof size too large and verification time too much.

sending $w$ to $V$, and sending an IP to verify $w$ satisfies $C$. Proof size might still be too big.