vkSplatting 2026.2.9 Cuts RTX Payload, Runs on AMD RADV

Michael Rubloff

NVIDIA has released vkSplatting 2026.2.9, which trims the ray tracing payload of its Vulkan gaussian splatting renderer and documents a build path for AMD GPUs under RADV. The tag is the first since 2026.2.0 replaced Phong shading with GGX PBR materials and added path traced lighting. The renderer added 3DGUT support and SPZ loading in earlier releases.
With a single splat set loaded, RTX_HAS_PARTICLES now encodes the splat descriptor count, and the per sample splatSetIdx array and currentSplatSetInstance leave the ray payload, saving PARTICLES_SPP + 1 dwords. The stochastic any hit shader now evaluates opacity only, the color and normal fields leave the payload, another 7 dwords, and spherical harmonics radiance is fetched once per ray in the raygen shader instead of once per candidate hit.
The release also fixes splat materials in the raster and hybrid paths with lighting enabled. In front to back mode every fragment wrote the per pixel splat ID to a color attachment with no depth test to arbitrate, so the farthest splat won.
Projects saved on Windows previously stored backslashes and failed to load on Linux, including the shipped samples once re-saved. Files up to version 7 still open.
The AMD path arrived through pull request 33 from ZJLi2013, who ran the raster, hardware ray tracing and hybrid pipelines on a Radeon AI PRO R9700. Two of its three commits landed on main under the contributor's authorship, a CMake fix that builds without the NVIDIA NGX SDK and a guard on --saveImage before the first headless frame, and the pull request was closed.
A second contribution, from Eason-Ruan of Manifold-ai, skips a default mesh pipeline compile at startup when --inputProject is set and a recompile during shutdown, and landed the same way, a commit on main and a closed pull request. The sample projects now feature winter garden content and a new video on lighting a scene from a radiance field.
Binaries for Linux and Windows are available now on GitHub.
New to Gaussian Splatting? Start here




