
Maciej Halber
509 posts

Maciej Halber
@maciejhalber
Computer Vision during the day, Graphics at night. He/him.


I've started writing my "game engine" in late 2001, when I began university. At the time, it used the then-modern OpenGL 1.1, following the famous NeHe tutorial site. Later, I rewrote it to use D3D9 and D3D10. As a Linux aficionado, I stuck with OpenGL for a long time because I found the API nimble and flexible, just like the operating system. OpenGL was not the best in general terms lately and did not advance the way D3D did, but it was good enough to naturally follow the modernization path started by AMD and DICE with Mantle and finally Vulkan. In a major rewrite in 2016, I moved my renderer to Vulkan. The next year, I had the opportunity to learn Vulkan and Metal best practices by porting the Serious Engine (The Talos Principle) to Apple. I never fully completed it due to time constraints, although I was pretty much done with the renderer, and I really grew to like the Metal API ideas. There is something elegant and beautiful in how Apple designs their APIs. Reading about legendary developer @SebAaltonen's adventures on HypeHype and seeing how cool the Metal debugger is gave me some FOMO. I upgraded to a new Apple device with M5, and since last year, in parallel with the previous AGC and D3D12 modernization to a GPU-first design, I've also been porting my renderer to Apple Silicon and the latest Metal. It is very impressive what Apple has achieved in a space dominated by the RGB trio, and I see Apple could become relevant as a gaming platform outside of mobile. Metal 4 can be used with pure C++ and has first-class citizen support from Apple, which makes the process very pleasant. The Metal debugger is awesome to use, and I'm having a blast just getting started.



CAPTCHA for 3D vision people





CAPTCHA for 3D vision people







It's funny that AI engineers are re-discovering all the graphics programming tricks. I remember reading the NeRF and Gaussian Splat papers long time ago, and found a lot of inefficiencies that we fixed years ago. Later they have been optimizing the techniques.


Wow is this the next WAN? If this actually runs on a consumer GPU, will change everything. But i fear it's yet another tech demo that "only need 10 H100 GPUs" to work like the video.











