Abstract: The dense conglomerate reservoir in the Mahu Depression is characterized by high conglomerate content, uneven particle size distribution and strong non-homogeneity, and the fracturing fracture extension mechanism is extremely complex, so the conventional physical model is limited by scale and boundary effects, which makes it difficult to realistically restore the fracturing construction conditions of deep conglomerate reservoirs. To address this problem, this paper relies on a super-large true triaxial hydraulic fracturing physical simulation system, selects a large conglomerate specimen with a size of 2.0 m×2.0 m×1.0 m, and adopts a real fracturing pumping capacity, and the simulated wellbore and fracturing materials are consistent with the on-site construction, so as to systematically investigate the effects of different conglomerate contents and particle sizes on the fracture extension law, and to realize the physical simulation of fracturing on an industrial scale. The experimental results show that gravel content and grain size have a coupling control effect on fracture morphology: When the content increases from 40% to 60%, the complexity of fracture network shows a trend of "decreasing and then increasing", and the main fracture deflection and branching development are mainly dominated by the spatial non-homogeneity of gravels; and the increase in the grain size makes the tortuosity of fracture show a trend of "decreasing and then decreasing". The increase of grain size makes the fracture tortuosity show the trend of "increasing and then decreasing", in which the medium grain size has the strongest shielding effect, forcing the fracture to expand around the gravel and form a complex branch network. The fracturing pressure curves further reveal that the expansion resistance is high under low content/small grain size conditions, and the curves are of step-shock type; whereas the main fracture paths are relatively regular under high content/large grain size conditions, and the pressure curves tend to be single-peak characteristics. The results reveal the dynamic network formation mechanism under the multi-scale mechanical action of gravel-matrix, and provide data support for the optimization of fracturing parameters in Makuhu tight conglomerate reservoir.