This paper applies ultra-high frequency data particularly tick-by-tick data from three precious metals, namely gold, silver and platinum to estimate realized volatility, and model HAR-RV models evaluate Value-at-Risk. The tick-by-tick data from January 1, 2011 to December 8, 2016 covering 2,070,720 of tick data per year with the total days of 1,439 days were used. The medRV is used to calculate the daily realized volatility and forwardly establish and compare HAR-RV in different functional form. The empirical results show that the minimum VaR and daily capital loss of precious metal are gold and follow by silver and platinum respectively.