VkSplat: High-Performance 3DGS Training in Vulkan Compute
This work addresses performance and compatibility bottlenecks in 3DGS training for GPU vendors beyond NVIDIA.
VkSplat implements a fully Vulkan-based 3D Gaussian Splatting training pipeline, achieving 3.3× speedup and 33% VRAM reduction over CUDA+PyTorch baselines while maintaining quality and cross-vendor compatibility.
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 $3.3\times$ speed and $33\%$ 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}