The rapid rise of 3D Gaussian Splatting (3DGS) has revolutionized computer graphics, replacing slow volumetric neural radiance fields with millions of real-time rasterizable 3D gaussians.
However, training dense spatial scenes from drone photogrammetry or high-resolution camera rigs requires immense memory bandwidth and sustained GPU boost clocks that quickly throttle laptops and thermal-limited local workstations.
NovaPod's high-density RTX 4090 and RTX 5090 instances provide 24 GB to 32 GB of dedicated high-speed GDDR6X/GDDR7 VRAM, allowing artists and spatial engineers to reconstruct complex city-scale scenes in under six minutes.
With pre-configured Unreal Engine Pixel Streaming templates and hardware NVENC encoding, interactive 4K 60fps viewports can be streamed directly to web browsers worldwide with under 16ms of glass-to-glass latency.
np launch \
--gpu RTX_5090 \
--template novapod/unreal-pixel-streaming:UE5.5 \
--disk 120GB \
--name spatial-gaussian-renderer