We present ABCD (Alpha-Composited Block Coordinate Descent), an out-of-core training framework for alpha-composited radiance fields, instantiated here for 3D Gaussian Splatting. Our method reformulates training as block coordinate descent over spatial partitions: only one block of parameters is active at a time, while all others are frozen. By exploiting the associativity of alpha blending, these inactive regions can be pre-rendered and collapsed into foreground and background RGBA images. As a result, for fixed partition size and image resolution, peak VRAM becomes O(1) with respect to total scene extent, rather than growing with full scene size. This enables GPUs with limited memory to train scenes that would otherwise not fit in core. In experiments, our method closely preserves the reconstruction quality of 3DGS, with less than 5% PSNR degradation, while ABCD with compositing ablated suffers roughly 40% degradation. Our code can be found at https://github.com/shiukaheng/abcd
Figures & tables
Figure 1. Overview of composited block descent with inactive regions cached as foreground and background RGBA images. Diagram showing an active spatial partition rendered between cached foreground and background image layers.
Figure 2. Qualitative comparison on a held-out view. ABCD with compositing ablated introduces severe boundary and consistency artifacts, while the full ABCD method closely matches the visual quality of 3DGS. Three rendered views comparing 3D Gaussian Splatting, ABCD with compositing ablated, and ABCD.
Figure 3. Training dynamics of 3DGS, ABCD with compositing ablated, and ABCD. Left: ABCD follows a loss trajectory similar to 3DGS, whereas removing compositing substantially impairs optimization. Right: ABCD retains the low active Gaussian residency of partitioned training, demonstrating that it preserves the memory behaviour of ABCD with compositing ablated while recovering most of the lost reconstruction quality. Training losses and active Gaussian counts for 3DGS, ABCD with compositing ablated, and ABCD.
Dataset
Method
VRAM (GB) ↓
RAM (GB) ↓
Time (s) ↓
PSNR ↑
SSIM ↑
Garden
3DGS
0.88
7.88
99.6
24.01
0.596
ABCD (ablated)
0.69
8.98
402.7
16.56
0.356
ABCD
0.80
25.02
1081.3
23.65
0.575
Kitchen
3DGS
0.97
3.73
67.4
27.03
0.861
ABCD (ablated)
0.75
5.60
441.3
15.64
0.504
ABCD
0.77
17.73
1041.0
25.80
0.839
Table 1. Comparison across methods on garden and kitchen scenes.
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× 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: https://github.com/harry7557558/vksplat
Training 3D Gaussian Splatting (3DGS) at billion-primitive scale is fundamentally memory-bound: each Gaussian primitive carries a large attribute vector, and the aggregate parameter table quickly exceeds GPU capacity, limiting prior systems to tens of millions of Gaussians on commodity single-GPU hardware. We observe that 3DGS training is inherently sparse and trajectory-conditioned: each iteration activates only the Gaussians visible from the current camera batch, so GPU memory can serve as a working-set cache rather than a persistent parameter store. Building on this insight, we introduce TideGS, an out-of-core training framework that manages parameters across an SSD-CPU-GPU hierarchy via three synergistic techniques: block-virtualized geometry for SSD-aligned spatial locality, a hierarchical asynchronous pipeline to overlap I/O with computation, and trajectory-adaptive differential streaming that transfers only incremental working-set deltas between iterations. Experiments show that TideGS enables training with over one billion Gaussians on a single 24 GB GPU while achieving the best reconstruction quality among evaluated single-GPU baselines on large-scale scenes, scaling beyond prior out-of-core baselines (e.g., approximately 100M Gaussians) and standard in-memory training (e.g., approximately 11M Gaussians).
Chonghao Zhong, Linfeng Shi, Hua Chen +4
Hong Kong University of Science and Technology · Great Wall Motor · Tsinghua University +1
3D Gaussian Splatting (3DGS) achieves remarkable novel view synthesis quality with real-time rendering, yet suffers from excessive memory consumption due to millions of Gaussian primitives. Existing pruning methods rely on heuristic importance scores or synchronous batch updates, leading to suboptimal compression and training instability. We propose VEDAL, a principled framework that formulates Gaussian pruning as variational free energy minimization. Our approach introduces (1) a prediction-error gating mechanism that asynchronously activates pruning based on per-Gaussian reconstruction uncertainty, and (2) a variational uncertainty head that models pruning decisions as latent variables with learnable priors. The free energy objective naturally balances reconstruction fidelity against model complexity through an information-theoretic lens. Extensive experiments on Mip-NeRF 360, Tanks&Temples, and Deep Blending demonstrate that VEDAL achieves 5.2x compression with only 0.31 dB PSNR drop, outperforming PUP 3D-GS by +0.05 dB at a higher compression ratio and LightGaussian by +0.35 dB at comparable quality, while maintaining real-time rendering at 185 FPS.
Aoduo Li, Jiancheng Li, Huan Ye +5
Guangdong University of Technology · Huizhou Boluo Power Supply Bureau, Guangdong Power Grid Co., Ltd. · Shenzhen Polytechnic University +1