Excited to see the ML technology I've been working on being released 😊

New short blog post:

"Running AutoGrad on GPU"

aschrein.github.io/jekyll/updat...

New short blog post:

"Image encoding using AutoGrad"

aschrein.github.io/jekyll/updat...

New short blog post:

"Adding MLP support to AutoGrad"

aschrein.github.io/jekyll/updat...

New short blog post:
"Adding tensor support to the toy AutoGrad"
aschrein.github.io/jekyll/updat...

New short blog post:
"Adding vector support to the toy AutoGrad"
aschrein.github.io/jekyll/updat...

New short post:
"Notes on AutoGrad"
aschrein.github.io/jekyll/updat...

New short post:
"Thinking about convolutions for graphics"
aschrein.github.io/jekyll/updat...

Sharing this job opportunity at my team: development and productization of ML+Graphics technologies for games. EU remote. careers.amd.com/careers-home...

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...

All HPG 2025 papers are now available on the EG digital library.
CGF papers: diglib.eg.org/handle/10.23...
Conference papers: diglib.eg.org/handle/10.23...
It has been a huge honor to act as HPG papers chair and guest editor and I look forward to an exciting program in Copenhagen.

Working on the next generation frontier technologies for denoising, supersampling, frame generation, radiance caching, GI, path tracing etc. etc.

Sharing this job opportunity at my team: development and productization of ML+Graphics technologies for games. EU remote. careers.amd.com/careers-home...

I wrote a blog post about mipmap level selection. pema.dev/2025/05/09/m...

A screenshot of RenderDoc where one can see calls to the WebGPU API

Who would be interested in a version of #RenderDoc that captures and replays calls to the #WebGPU API (rather than calls to the underlying DirectX/Vulkan/Metal API)?

This is an early test that only lists the API calls, but already promissing! Will share when usable.

Shadertoy

TIL.
I've been using Moeller-Trumbore's ray/tri intersection almost everywhere, for ages. It includes several culling tests, including: u<0, v<0, u>1, u+v>1.
Then I encountered this Shadertoy yesterday:
www.shadertoy.com/view/wlsfRs
This one skips the test for u>1.
Which totally makes sense now.

wow looks awesome! there's actually quite a bit of free space left even. insane!

@fcama.bsky.social and the other wizards from the FSR team should be very proud!

AMD's machine learning powered FSR4 on RNDA4 is a monumental leap forward for its GPU division. FSR4 thoroughly obliterates FSR3 making it so that having an AMD GPU no longer means compromising on IQ. But how does it compare to DLSS old and new? Find out in my video!
youtu.be/nzomNQaPFSk

Graphics Programming weekly - Issue 381 - March 2nd, 2025 www.jendrikillner.com/post/graphic...

ML in gfx is quite a switch, you need to align a lot of figures to begin getting it to work. There's a clear percolation threshold - a lot of the times it's unclear it should work at all, and on top of that there's not a ton of papers out there(for realtime), almost tribal knowledge at this point.

With the cooperative matrices starting to standardise I'm kind of starting to wish for a cross-studio cooperation to see an open ML upscaler good enough to kill proprietary solutions.
I don't have the power to start something like that unfortunately.

target is patches from the noisy sampled image, similar to the image compression

different input encoding

Same MLP applied in world space to learn lighting

Ported my old WebGPU MLP image compressor toy to my new d3d12 framework aschrein.github.io/mlp_compress...

I wrote about the journey to make GPU texture compression work everywhere, including some history on how I got into this field. Check it out!

www.ludicon.com/castano/blog...

Got basic soft direct shadows going with blue noise eheitzresearch.wordpress.com/762-2/

Enabled AGS for the shader clock extension