Saved in:
Bibliographic Details
Main Authors: Chen, Jingxiang, Ibrahim, Mohamed, Liu, Yang
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.00219
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913079735877632
author Chen, Jingxiang
Ibrahim, Mohamed
Liu, Yang
author_facet Chen, Jingxiang
Ibrahim, Mohamed
Liu, Yang
contents 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}
format Preprint
id arxiv_https___arxiv_org_abs_2605_00219
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle VkSplat: High-Performance 3DGS Training in Vulkan Compute
Chen, Jingxiang
Ibrahim, Mohamed
Liu, Yang
Computer Vision and Pattern Recognition
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}
title VkSplat: High-Performance 3DGS Training in Vulkan Compute
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2605.00219