Another #ThreeJS PR for today: Specular Occlusion ("GTSO") via GTAO. Improved it slightly to avoid dark outlines on objects.

https://github.com/mrdoob/three.js/pull/35052

Today's #ThreeJS PR - Specular Reflection Probes - For fast specular reflections on low end devices. Per pixel, box-projected, and build on WebGPU's cube-array textures. https://github.com/mrdoob/three.js/pull/35046

Today's #ThreeJS PR: Added native 16-bit linear float HDR loading using browser-based Jpeg XL decoding, float16 Canvas and globalHDRheadroom features.

https://github.com/mrdoob/three.js/pull/35015

Today's ThreeJS contribution: Real-time oriented UV unwrapped for ThreeJS native meshes and scenes. 10x smaller and 3x faster than xAtlas on average.

https://github.com/mrdoob/three.js/pull/34944

Adding support for priority-based loading to ThreeJS's FileLoader - prioritize your main assets first { priority: 'high' }, load LODs and background elements later { priority: 'low' }

https://github.com/mrdoob/three.js/pull/34936

Global Illumination Light Probes in @ThreeJS is now 4x faster to calculate and more accurate. The calculation was CPU-bound, so I used an ArrayCamera to bulk render probes, and I integrate their results fully, rather than just 512 samples.

https://github.com/mrdoob/three.js/pull/34909

Bug report: https://issuetracker.google.com/issues/571218887

Proposed fix: https://swiftshader-review.googlesource.com/c/SwiftShader/+/77868

Nasty surprise while working with #WebGL today. Turns out Chrome's SwiftShader backend has a buggy atan2 implementation when runtime y = negative zero.

https://github.com/mrdoob/three.js/pull/34913
https://github.com/bhouston/swift-shader-atan-bug

New project: Fidelity test suite for @garrettkjohnson's three-gpu-pathtracer vs @blender.org . Compares both the legacy #WebGL and new #WebGPU renderers.

https://bhouston.github.io/three-gpu-pathtracer-fidelity/?ref=blender&renderers=webgpu-new

https://github.com/bhouston/three-gpu-pathtracer-fidelity

OpenDLSS 5 comes to ThreeJS. NVIDIA’s DLSS 5 neural rendering network, as @threejs compute nodes, byte for byte. https://github.com/bhouston/three-dlss-nr/

Definitely don’t want this on your lap. But sustained CPU and GPU no longer throttles!

Added thermal paste to the MacBook Air to reduce thermal throttling under sustained load. It no longer throttles under sustained 100% GPU load, the bottom surface though is 124 F/50 C.

"git-dedup" - a Git wrapper that makes git faster and smaller for agentic workflows. Specifically checkouts and submodules can be slow if you have agent fleets on the go. This fixes that:

https://ben3d.ca/blog/git-dedup-faster-smaller-checkouts

I love my MacBook Air 95% of the time except when I have to do a bunch of Blender renders. It throttles down since it is at 100% sustained usage and becomes ultra slow.

Want to inspect #MaterialX materials in your coding IDE? Try my new #ThreeJS-based MaterialX Viewer extension:

https://marketplace.visualstudio.com/items?itemName=benhouston3d.mtlx-vscode-extension

https://open-vsx.org/extension/benhouston3d/mtlx-vscode-extension

2.4x Faster Native #GPU Testing for #Vitest and #Jest without a Browser.

Thanks to Renaud Rohlinger for the inspiration (and this is partially built on his node-webgl package.)

https://ben3d.ca/blog/native-gpu-testing-for-vitest-and-jest

Want to write simpler & faster tests your #WebGPU or #WebGL code in @vitest ? I've created two Vitest environments just for you! https://github.com/bhouston/vitest-gpu

I want developers to be able to write performant #WebGPU code. To do that, they need to know what different operations, code patterns, and optional features actually cost on the devices people use. https://ben3d.ca/blog/introducing-webgpu-bench

Neural Texture Compression (NTC) for #ThreeJS.

NTC shrinks material sizes by 90% by compressing all textures together into a single compact neural network.

This is also the first open-source NTC implementation.

Trainer + #ThreeJS runtime library here:
https://three-ntc.ben3d.ca

Web3D Survey's new GPU benchmarking page now has a sharable results URL. Here is mine:

https://web3dsurvey.com/benchmark/401e6467-b0d0-40a7-bc60-d0240f393500