GEMM-GS: Accelerating 3D Gaussian Splatting on Tensor Cores with GEMM-Compatible Blending

Paper Link

Haomin Li*, Bowen Zhu, Fangxin Liu*, Zongwu Wang, Xinran Liang, Li Jiang, Haibing Guan

This is the official implementation of the paper "GEMM-GS: Accelerating 3D Gaussian Splatting on Tensor Cores with GEMM-Compatible Blending" [DAC 2026]

Run

You can manually toggle whether GEMM-GS is enabled by changing bool use_gemm = true; at line 792 in GEMM-GS/submodules/gemm-gs/cuda_rasterizer/forward.cu.

Citation

@inproceedings{li2026GEMMGS,
  title={GEMM-GS: Accelerating 3D Gaussian Splatting on Tensor Cores with GEMM-Compatible Blending},
  author={Li, Haomin and Zhu, Bowen and Liu, Fangxin and Wang, Zongwu and Liang, Xinran and Jiang, Li and Guan, Haibing},
  booktitle={2026 63rd ACM/IEEE Design Automation Conference (DAC)},
  year={2026},
  organization={IEEE}
}

• Bowen Zhu: zhubowen@sjtu.edu.cn
• Haomin Li: haominli@sjtu.edu.cn
• Fangxin Liu: liufangxin@sjtu.edu.cn
• Li Jiang: ljiang_cs@sjtu.edu.cn