What to Learn to Be a Graphics Programmer(blog.demofox.org) |
What to Learn to Be a Graphics Programmer(blog.demofox.org) |
I started in 2001, when NVidias first Geforce 1 ("the Gigatexl shadercard") was first announced: The field developed since then with so much speed and innovations, it blows my mind of. Compared to what we could do 25years ago, the tech today is just fu*ing impressive.
Though, with this impressiveness comes a big "but": The space is developing at a speed which is really really scary. Nvidia came up with AI-based effects to influence scene & assets on their own - back then, we wouldnt have even thought about that this will be possible some day in realtime.
I do not know if its possible at all to be a "decent pro" in this field now - let me use other words: "Where is todays Jon Carmack?" - he was famous for squeezing everything out of the hardware, using ideas very hidden in the community etc. - today, there is not any competitive moat for people like him (he actually lives on his legacy), and that is because the field is so vast and evolving so fast that there is no chance to become the next one
So here’s another perspective. If all you have done is web apps and Kubernetes, for example, do get into graphics programming. The feedback cycle is exhilarating, and you get to appreciate how mind boggingly fast your average computer is. You’ll get to optimize things that are ultimately unimportant because you have never learned how quick things are at the low level. There are a ton of resources and the maths is not too bad. You might find that 3D modeling is a creative outlet you didn’t know you needed. Even if completely inapplicable to your day job, you’ll find new ways to appreciate the art of programming computers, and might just decide to never touch Kubernetes again and spend the next 5 years writing your own game engine in your spare time. There are a lot of crazy people like that, and the community of hobbyists that are not ground down by life and game dev as a career is larger than you’d think. The Graphics Programming discord is a welcoming place if you want to check it out.
Go for it!
That's not the argument being made here. The field is changing. I had a good career in graphics, my life wasn't wasted at all. That doesn't mean a college student would have the same experience starting today.
Given that almost everyone who wants to be a "graphics programmer" is also somehow gaming industry adjacent, it is extremely fair to ward them off from the folly. I do the same for anyone wishing to do "VLSI hardware engineering." If you have the skill to do either of those, you almost CERTAINLY have the skill to do something else that is almost as interesting and not saddled by garbage employers.
The primary problem with being a "graphics programmer" beyond a tyro is that the biggest consumer of graphics programmers is the game industry which is a notoriously shitty and wretched industry. Every ... single ... employer. So, from the point of view of future potential, "graphics programmer" has very little upside over pretty much ANY other type of programmers.
Second, "learning graphics programming" is like "learning phone programming", you spend more time fighting godawful software infrastructure more than you do actual programming. AI actually kind of helps this, but it doesn't completely remove the fact that 80% of your knowledge has a half-life of 18-24 months.
Finally, saying "I want to learn graphics programming" is like saying "I want to learn engineering." What "graphics programmer" means is vastly underspecified. 3D game rendering and 3D/2D CAD rendering and 2D vector rendering are completely different skillsets. GPUs are great at the first and kinda okay at the second and kinda lousy at the third. Which kind of "graphics programmer" are you even going to be?
The trick there is that they both have related physical analogs, and machine learning math doesn't really (in that while you can visualize them spatially, it doesn't seem to help solve any problems in that space.)
I've been in ML for ~5 years in multiple FAANGs and I have never seen a rotation matrix.
We see folks posting photo real, Gaussian splat FPS maps here every now and then but without also innovating on gameplay its just a tech demo. Those don't cut it these days.
You don’t need to be JC to earn a decent living as a graphics/game programmer.
I also have some notebooks with SymPy code examples here: https://github.com/minireference/noBSLAnotebooks
Graphics aren't my specialty, but as a musician, sound designer and producer, by far the most effective/influential audio DSP coders I'm aware of understand the basics of music, the physics/acoustics of sounds, and the gotchas at the interface between discrete digital processes and how we perceive and interpret stimuli.
Start with just <a-box> and <a-sky>, add some animations, then add some community components if it's not enough. Still not enough then modify via ThreeJS, all the way to shaders. A-Frame is amazing so thanks for creating and maintaining it!
PS: Oh, and you can even do AR and VR with it.
This guy has some good art: https://www.khanacademy.org/profile/kaid_1019042693170894950...
If you want to work with Windows, probably DirectX.
2. Make awesome shaders. Check this out: https://fragcoord.xyz
I would say being a long-time user of Photoshop and Blender helps a lot. It's not a main tool, but supplemental. Maybe AI will take over some of this though.
Hell, maybe that other stuff too, hahaha!
Once you have that intuition, the rest is all figuring out the stages of the graphics pipeline and the frameworks like opengl and their constituent data structures.
The most useful resources I've found for graphics are scratchapixel, UC Davis' graphics lectures, songho's articles, and Essential Math for Games and Interactive Applications. I highly recommend you read this last resource front to back. Seriously, its the best freaking math reference for graphics out there.
But knowing theory is not sufficient. You also need to get your hands dirty by writing code: learn how to build a software rasterizer (check out Tiny renderer) and a ray tracer (Ray Tracing in One Weekend series). Preferably in a language like C++. Then move onto APIs. I recommend you learn OpenGL, but if you're okay with being confused as all hell try Vulkan. Or WebGPU if you're a hipster (/s).
Finally, try to build some stuff. A simple engine. A non-trivial technique. A game. Whatever.
Unfortunately, you're unlikely to get hired working as a rendering engineer without having serious connections, or by having adjacent experience in the industry. Doubly so now that everyone is convinced junior engineers are unnecessary.
I will definitely check out Essential Math for Games and Interactive Applications, I feel I need some solid understanding of theory to see how it all fits together.
I think graphics programmers benefit from having an artistic mindset, but they usually work so low level that it isn’t necessary to be successful.
AI changed the calculus a bit (or at least, it has the potential to) but I think that was a huge part of the whole "learn to code" movement in the mid 2000s, to start treating software development as a "feature, not a product" of existing experts in their field so that the people most familiar with their domain are actually the ones making the software instead of having to translate the requirements down to a dev team.
They needed warm bodies for their projects, as the usual source of manpower was grinding leetcode to work on bigtech at salaries that would make an accenture business type vomit in disgust.
The rapid advances, in a trend replicating throughout society, push out the middle in favor of the top.
The field is always changing. You could find people in the 80s saying ‘I had a good career in graphics, a college student would not have the same experience starting today’
How much do you even think about explicit matrix math when doing high-level ML?
The whole ML field is basically about starting from random points and trying to find useful shapes and constraints. Basically like trying to get object likeness in clouds
Not so hypotheticals -- Heck the inputs that you want labelled could be rotation matrices. The desired output could be a rotation matrix. Generating more convenient features could be via a rotation matrix. Dimensionality reduction could be through a reduction matrix. Sparsity could be encouraged by proper use of rotation matrices. Shows up if you want to build in group theoretic invariance in your predictive model.
(*) If you consider Householders then even more