Moore Threads MTT S5000 GPU Supports ByteDance's Open-Source Protenix-v2 Model
Moore Threads announced that its flagship MTT S5000 GPU card has achieved full end-to-end inference support for ByteDance's open-source biomolecular structure prediction model, Protenix-v2. The model is capable of high-precision predictions for complex molecular structures, including proteins, DNA, RNA, and small-molecule ligands. This achievement marks the technical verification of Moore Threads' hardware across three key AI for Science (AI4S) scenarios: macromolecular structure prediction, genome modeling, and medical imaging. It strengthens China's domestic compute base for pharmaceutical enterprise research and scientific institutions seeking alternatives to Western GPU hardware. Protenix-v2 features 464 million parameters and matches or exceeds DeepMind's AlphaFold 3 on several key benchmarks, particularly in antibody-antigen complex predictions. The MTT S5000 GPU, built on Moore Threads' fourth-generation Pinghu architecture, provides the hardware foundation for running these heavy biomolecular inference workloads.
## BACKGROUND
Protenix-v2 is an open-source biomolecular modeling system developed by ByteDance's Seed team to accelerate drug discovery and biological research. Moore Threads is a Chinese semiconductor company focused on developing domestic general-purpose GPUs to support AI workloads and advance national technological self-reliance.