The Manin constant c of an elliptic curve E over Q is the nonzero integer that scales the pullback of a Néron differential under a minimal parametrization φ : X0(N ) Q ։ E into the differential ω f determined by the normalized newform f associated to E. Manin conjectured that c = ±1 for optimal parametrizations, and we prove that in general c | deg(φ) under a minor assumption at 2 and 3 that is not needed for cube-free N or for parametrizations by X1(N ) Q . Since c is supported at the additive reduction primes, which need not divide deg(φ), this improves the status of the Manin conjecture for many E. For the proof, we settle a part of the Manin conjecture by establishing an integrality property of ω f necessary for it to hold. We reduce the latter to p-adic bounds on denominators of the Fourier expansions of f at all the cusps of X0(N ) C and then use the recent basic identity for the p-adic Whittaker newform to establish stronger bounds in the more general setup of newforms of weight k on X0(N ). To overcome obstacles at 2 and 3, we analyze nondihedral supercuspidal representations of GL2(Q2) and exhibit new cases in which X0(N ) Z has rational singularities. E may be more canonical within the same isogeny class: for instance, X 1 (11) Q and X 0 (11) Q are distinct isogenous elliptic curves. The multiplicity one theorem ensures that the φ-pullback of a Néron differential ω E is a nonzero multiple of the differential ω f ∈ H 0 ((X Γ ) Q , Ω 1 ) associated to the normalized newform f whose Hecke eigenvalues agree with the Frobenius traces of E: