@SonnyBonds I think if you implement everything in compute and compile the shaders with strict IEEE floating point math, you should get reasonably deterministic results between platforms. Raster shaders can probably also be made to be pretty close, but it's defined loosely enough that it varies between vendor.
„The features matter a lot less than the people who are using the platform. […]
It can sometimes be a bit misleading when you get a lot of ideas and feature requests in a community, and the conversations become, ‘We definitely need feature X to grow because that’s what’s stopping people from using the platform.‘ While that’s true in some cases, the sad reality is that any flaw can be overlooked as long as the people you want to reach are there.“
(Un po' di) storia contemporanea.Nessuno si aspettava ChatGPT - Il Posthttps://www.ilpost.it/2025/11/30/chatgpt-tre-anni/> Quando tre anni fa OpenAI mise online il chatbot che ha cambiato il settore tecnologico prese tutti alla sprovvista, compresa sé stessa
@SonnyBonds I think if you implement everything in compute and compile the shaders with strict IEEE floating point math, you should get reasonably deterministic results between platforms. Raster shaders can probably also be made to be pretty close, but it's defined loosely enough that it varies between vendor.