For　the　complex　transient　thermo-hydraulic　restart　process　of　a　long-buried　hot　oil　pipeline,　it　takes　a　long　time　to　solve　the　restart　simulation　model,　which　cannot　meet　the　requirements　of　rapid　restart　prediction　and　assessment.　The　modern　GPU　is　specialized　for　compute-intensive　and　massively　parallel　computation,　and　it　has　great　potential　to　accelerate　transient　thermo-hydraulic　restart　simulation.　According　to　the　computational　characteristics　of　restart　model　solutions,　four　types　of　GPU　parallel　strategies,　from　the　mapping　of　grids　to　the　convergence　judgment　of　the　discrete　equation　solution,　are　combined　to　accelerate　the　whole　process　of　a　transient　thermo-hydraulic　restart　simulation.　The　results　indicate　that　the　total　computation　time　of　the　tran-sient　thermo-hydraulic　simulation　is　reduced　from　185.4　min　to　8.8　min,　corresponding　to　an　acceleration　rate　of　21.1,　which　exhibits　a　good　acceleration　effect　for　a　single　GPU　and　is　beneficial　to　the　rapid　assessment　of　restart　schemes.　Based　on　the　GPU-accelerated　numerical　results,　the　complex　transient　thermo-hydraulic　restart　pro-cess　with　the　changing　soil　temperature　field,　the　non-Newtonian　thixotropic　behavior　of　crude　oil　and　flow　pattern　evolution,　is　investigated.　The　analysis　reveals　that　the　dominant　influencing　factors　of　the　restart　process　vary　with　the　restart　time,　and　the　flow　pattern　evolution　plays　an　important　role　in　the　thermo-hydraulic　characteristics　of　the　restart　process.