English

VkSplat: High-Performance 3DGS Training in Vulkan Compute

Computer Vision and Pattern Recognition 2026-05-04 v1

Abstract

We present VkSplat, a high-performance, cross-vendor 3D Gaussian Splatting (3DGS) training pipeline implemented fully in Vulkan compute, addressing performance and compatibility limitation of existing training pipelines. With various optimizations, we achieve 3.3×3.3\times speed and 33%33\% VRAM reduction over CUDA+PyTorch baseline, maintaining quality, and demonstrating compatibility across GPU vendors. To the best of our knowledge, this is the first fully-Vulkan-based 3DGS training pipeline that achieves state-of-the-art performance. Code: \href{https://github.com/harry7557558/vksplat}{https://github.com/harry7557558/vksplat}

Cite

@article{arxiv.2605.00219,
  title  = {VkSplat: High-Performance 3DGS Training in Vulkan Compute},
  author = {Jingxiang Chen and Mohamed Ibrahim and Yang Liu},
  journal= {arXiv preprint arXiv:2605.00219},
  year   = {2026}
}

Comments

Submitted to Eurographics 2026 - Short Papers

R2 v1 2026-07-01T12:44:30.971Z