Puget Systems spotlights NVIDIA Blackwell AI workflows for production at SIGGRAPH

The transition of AI from a post-production research curiosity to a real-time production-ready asset is accelerating. By integrating 4D Gaussian Splatting and AI relighting directly into Unreal Engine, Puget Systems is lowering the barrier for high-fidelity virtual production environments. This shift reflects a broader streaming ecosystem move toward 'neural graphics,' where physically accurate lighting and environments are generated via inference rather than manual modeling. For the industry, the metric to watch is the throughput of RTX Neural Shaders in handling complex 3DGS scenes, which could significantly reduce the cost and latency of volumetric streaming and immersive storytelling.

The professional hardware landscape has shifted significantly following the March 2025 launch of NVIDIA’s Blackwell workstation lineup. According to Puget Systems’ own internal performance benchmarks from July 2026, the RTX PRO 6000 Blackwell delivers an average 48% speed increase in GPU rendering tasks like Blender Cycles compared to the previous Ada Lovelace generation. The card's 96GB of GDDR7 VRAM is specifically designed to address 'Out of Memory' bottlenecks commonly encountered in 4K virtual production and Large Language Model (LLM) fine-tuning. However, the high 600W power requirement of the Workstation Edition has necessitated specialized thermal designs, leading to the rise of the 300W 'Max-Q' variant for high-density multi-GPU clusters, per NVIDIA’s technical documentation from early 2026.

In the software layer, 3D Gaussian Splatting (3DGS) has moved beyond static capture into dynamic 4D applications. As reported by Volinga in July 2026, their latest plugin for Unreal Engine 5.8 adds native support for sub-level streaming and DLSS 4 to maintain fluid frame rates in dense volumetric scenes. This matches a broader industry trend toward AI-augmented production tools, exemplified by Corridor Digital’s adoption of custom Unreal plugins to automate lighting tasks that previously required manual PBR material creation. Furthermore, at SIGGRAPH 2026, major cloud providers like AWS announced the expanded availability of EC2 G7e instances, which provide virtualized access to these same Blackwell Server Edition GPUs, enabling studios to scale neural rendering without the capital expenditure of localized $13,000 graphics cards.