This paper investigates the sensing performance of two intelligent reflecting surface (IRS)-enabled non-line-of-sight (NLoS) sensing systems with fully-passive and semi-passive IRSs, respectively. In particular, we consider a fundamental setup with one base station (BS), one uniform linear array (ULA) IRS, and one point target in the NLoS region of the BS. Accordingly, we analyze the sensing signal-to-noise ratio (SNR) performance for a target detection scenario and the estimation Cram\'er-Rao bound (CRB) performance for a target's direction-of-arrival (DoA) estimation scenario, in cases where the transmit beamforming at the BS and the reflective beamforming at the IRS are jointly optimized. First, for the target detection scenario, we characterize the maximum sensing SNR when the BS-IRS channels are line-of-sight (LoS) and Rayleigh fading, respectively. It is revealed that when the number of reflecting elements $N$ equipped at the IRS becomes sufficiently large, the maximum sensing SNR increases proportionally to $N^2$ for the semi-passive-IRS sensing system, but proportionally to $N^4$ for the fully-passive-IRS counterpart. Then, for the target's DoA estimation scenario, we analyze the minimum CRB performance when the BS-IRS channel follows Rayleigh fading. Specifically, when $N$ grows, the minimum CRB decreases inversely proportionally to $N^4$ and $N^6$ for the semi-passive and fully-passive-IRS sensing systems, respectively. Finally, numerical results are presented to corroborate our analysis across various transmit and reflective beamforming design schemes under general channel setups. It is shown that the fully-passive-IRS sensing system outperforms the semi-passive counterpart when $N$ exceeds a certain threshold. This advantage is attributed to the additional reflective beamforming gain in the IRS-BS path, which efficiently compensates for the path loss for a large $N$.