Details how it was created can be found in the GPU Zen 4 book!

GitHub - boksajak/MLPZen Contribute to boksajak/MLPZen development by creating an account on GitHub.

A gentle reminder my MLPZen project is up on GitHub. You get:
- MLP inference and training in standard HLSL, in-line anywhere in your shaders
- 2D and 3D Hash-grid
- Adam optimizer
- Dx12 sample app
- Link: github.com/boksajak/MLP...
Looking forward to how people will use this!

If Aai hallucinator gizmos kill tech like this, I will be infinitely sad

Workfls for me now, if ot does not work for you, the paper is also available on arxiv

Love this ray tracing statue 🕺

features 4 bounces of light, denoiser, physically-based lighting and 1000FPS

I made a game! Do you like playing your favorite arcades but feel like there's not enough ray tracing?

You can now play my "bin packing" classic with full path tracing.

Open source and free (executables in the bin folder):
github.com/boksajak/Bin...

My MLPZen project comes with a sample showing a simple neural texture compression (NTC). You don't need a complex vendor-locked SDK with proprietary extensions to start experimenting with NTC for your game. This project is pure DirectX 12/HLSL. Github: github.com/boksajak/MLP...

Hot off the press, the GPU Zen 4! You can get yours on Amazon: www.amazon.de/GPU-Zen-Adva...

I am happy to announce publication of my chapter titled "Real-Time Neural Network Implementation for GPUs" in GPU Zen 4 book, available now.

Chapter comes with full MIT-licensed source code, available on Github here: github.com/boksajak/MLP...

The book is available on Amazon.

Forgot to post it here, but I recently went through @boksajak.bsky.social's great Crash Course in Deep Learning. To make sure I understood everything, I implemented it in Odin.

This straightforward CPU version is very slow as you might expect, but also easy to follow.

github.com/wookie41/odi...

With interior? Is there an art gallery inside?

This is a great article about ReGIR / World-Space ReSTIR with nice improvements by Tom Clabault:
tomclabault.github.io/blog/2025/re...

Publications - AMD GPUOpen Discover our published publications.

See what our research teams have been up to with our latest papers:

📝 Joint Denoising and Upscaling via Multi-branch Network.
📝 Cardioid Caustics Generation with Diffusion Models.
📝 Geometric Integration for Neural Control Variates.

📥 View here: gpuopen.com/learn/public...

New blog post! Behind the scenes of some of the techniques involved in making our last PC demo 💫 gboisse.github.io/posts/this-i...

I love to plqy games with ambient occlusion only 🙉

One group is stochastic friendly to ray tracing while the other group is more suited to rasterization pipeline. I agree though that over time, stochastic RT approaches will universally be the high end option

Graphics Programming weekly - Issue 413 - October 19th, 2025 www.jendrikillner.com/post/graphic...

I have been coding in C++ for 26 years, but I admit I cannot read what this does 🤨

A comparison between AMD Neural Light Sampling and ReSTIR.

Get up to speed with our latest advanced graphics research:

📝 GATE (Geometry-Aware Trained Encoding) for neural rendering.
📝 Real-time procedural GPU tree generation.
📝 Neural Visibility Cache for real-time light sampling.

📥 Download the papers: gpuopen.com/learn/public...

Our "Neural Visibility Cache" paper has been published in JCGT! Here is a link: www.jcgt.org/published/00...

I found it cool to hear the motivation for High-Frequency Learnable Encoding for NRC\NIRC\Neural Ambient Occlusion from the perspective of Kernel Machines!

Classics 😊

I will be presenting our "GATE: Geometry-Aware Trained Encoding" at HPG on Wednesday at 9:50AM. See you all in Copenhagen soon! Paper link is here: diglib.eg.org/bitstream/ha...

I'll buy your book

This is the kind of story I needed to hear today 🙂. The power of mexican cuisine 💪

Just a good balance for performance. Feel free to try larger networks, but tou may need to increase number of features per level of the hashgrid as well

(1) Yes, we mention that as "clustering" for future work
(2) We could have a fixed MLP serving as a decoder, or non-neural function reading stuff stored on geometry with GATE.

New papers day! Preprints of our „GATE: Geometry-Aware Trained Encoding“ to be presented at HPG and „Neural Visibility Cache for Real-Time Light Sampling“ currently on Arxiv are available here: boksajak.github.io/blog/Preprin... Kudos to co-authors Daniel Meister and Carsten Benthin. Enjoy!

Lol, and you need to use the advance lever to move the "film" to the next frame. Perfect for people who want the inconvenience of using film with none of the benefits. I am glad it exists and that I don't have to use it.

You're within the order of magnitude of quarter res still. We can agree