Characterisation of cooler atmospheres of super-Earths and Neptune sized objects is often thwarted by the presence of clouds, hazes and aerosols which effectively flatten the transmission spectra. High-Resolution Spectroscopy (HRS) presents an opportunity to overcome this limitation by having the ability to detect molecular species whose spectral line cores extend above the level of clouds in these atmospheres. We analyse High-Resolution infrared observations of the warm Neptune GJ 3470b taken over two transits using CARMENES (R $/sim$ 80400) and look for signatures of H$_{2}$O in these transits with a custom pipeline fully accounting for the effects of data cleaning on a potential exoplanet signal. We find that our data is able to weakly detect an injected signal equivalent to the best-fit model from previous HST WFC3+Spitzer observations. However, we do not measure any significant detection using the actual observations. Using a Bayesian retrieval tool on the two observed transits to put simultaneous constraints on the abundance of H$_{2}$O and the cloud top-deck pressure selects for a family of degenerate models, which spans from very high abundance (log$_{10}$(H$_2$O) = -0.5) and cloud-free models (highly compressed heavy atmospheres), to super-solar abundances (-2.5$<$log$_{10}$(H$_2$O)$\leq$-1) at high cloud deck models (-2.5$>$log$_{10}$(P)$\geq$-4.5), and then to slightly super-solar and sub-solar abundances (-4.5$<$log$_{10}$(H$_2$O)$\leq$-2.5) with moderately high cloud deck models (-1.5$>$log$_{10}$(P)$\geq$-2.5). This is a broader range compared to, but is also compatible with published results from low resolution at a 1$\sigma$ level.