Alex Fuller

Professional 3D artist guy, hobbyist games/graphics developer. Something something my opinions are my own and such.

[Alex Fuller @boberfly]

@HealthRanger Good timing Tenstorrent just released their Galaxy Blackhole - wish I could afford it, but the AMD Strix Halo I bought last year was a good idea (despite the firmware only working after Jan 2026 or so)

[Alex Fuller @boberfly]

@BGatesIsaPyscho @clif_high Bro discovered caustics

[Alex Fuller @boberfly]

@SebAaltonen @ThreatInteract Probably why Vulkan never got that outside of when NV Turing brought in VK_NV_shader_image_footprint also it's a whack-a-mole on native dx12 hw https://t.co/GaB3GBD1b8

[Alex Fuller @boberfly]

@MatrixQdeBreakr @Omazonn @RamboAndFrens @NewsTreason @Spaceshot76 🌕👻⚡️

[Alex Fuller @boberfly]

Yep...

[Alex Fuller @boberfly]

@stefan_3d @yiningkarlli Just started to dig into spirv because I am wanting OSL on all backends and not just SVM - seems like as of LLVM 21 or so the spirv backend can target Intel and amdhsa targets and relocatable code to make it work, which I think specialisation constants could be done too?

[Alex Fuller @boberfly]

@MCH2024 @SebAaltonen Sure enough they split these into smaller and separate kernels and don't bake multiple megakernel due to the raster visibility buffer and generated gbuffer after that

[Alex Fuller @boberfly]

@MCH2024 @SebAaltonen Reading GPU Zen 4 about Doom TDA I was quite impressed at how they manage material variety this time they can achieve what you'd normally see in a standard offline path-tracer (2 spec lobes, SSS, dedicated hair shader) and up to 8 layer inputs baked at runtime

[Eric Lengyel @EricLengyel]

New blog post: A Decade of Slug
This talks about the evolution of the Slug font rendering algorithm, and it includes an exciting announcement: The patent has been dedicated to the public domain.
https://t.co/xWEz0q2c4N

[Alex Fuller @boberfly]

https://t.co/PnZg0Ps6K6 nice! (also Brecht landed something I worked on awhile back for Cycles but merged the two together - true corner/face-varying normals that are octahedrally compressed at 3:1 ratio without visual difference)

[Alex Fuller @boberfly]

https://t.co/hLxdBP1zAc weird, I was last night doing a deep-dive on how you'd emulate clusters/CLAS by just making tiny HPLOC-built BVHs into the AABB mode of BLAS with a special hit shader, but if the driver will do it vendor-neutral and on older hardware then cool too...

[Alex Fuller @boberfly]

@VadimYuryev @jimmyjames_tech https://t.co/ke1AZ02aK7 yep pretty good slightly better than a 4090 laptop and almost as good as the M3 Ultra, amazing stuff

[Alex Fuller @boberfly]

@VadimYuryev @jimmyjames_tech Keen to see blender open-data on this, my prediction is around 4090 laptop version for the 40-core variant but let's see...

[Alex Fuller @boberfly]

@HarleyRyder33 @goddek I'll confidently warp spacetime for ya and implant some moving clouds on cg textures next time

[Alex Fuller @boberfly]

@HarleyRyder33 @goddek Dunno man in compositing software like Nuke you could make some noise movement on the original photos from cg textures as well...

[Alex Fuller @boberfly]

@admiral5137 @goddek Yeah that story is hard to believe imo - but I did hear that too, I feel it is a huge stretch vs just a vfx piece - sorry...

[Alex Fuller @boberfly]

@chrismartenson @goddek All roads lead to ancient Phrygia, the opening ceremony in Paris screamed Phrygian...

[Alex Fuller @boberfly]

@beyond_fps Also for AMD they've finally landed proper callable functions/shaders and have a better BVH construction so I'm expecting RT to be a lot better soon in Mesa3D 26.x series *shrugs*

[Alex Fuller @boberfly]

@beyond_fps VK_EXT_descriptor_heap will fix this - getting DX12 on Vulkan right now is suboptimal on Nvidia hardware while it's ok with AMD more or less due to the underlying architecture that can map well to the current translation technique, as well as some other extensions RADV supports