pouët.net

Or perhaps the engine could render in original Doom quality, and AI pass enhancing rendered frames to UE5/photorealistic quality. That way you could ditch local powerful GPUs and CPUs and use a Pentium 1 level machine. Of course with 5G connection to MS AI server farm that powers the AI enhance pass. That actually sounds just about stupid and wasteful enough to make it happen by the AI crowd. First as a conceptual masturbation, then commercially. It makes all those people who artistically perfected render engines in the last 30 years redundant going forward. Wow, right? Yay, even. 🤮

congrats, you just invented something that nvidia invented many years ago!

(with that.. has anyone already tried RTX Remix on some late 2000s flyby? :) )

Oh, but it didn’t wear the AI badge. Perhaps because the farm was doing the actual, you know, rendering, not AI enhancing. But whatever.

I believe RTX Remix only works with D3D9 (not OpenGL, not D3D8), and only fixed function… not sure how many demos fit that bill? I mean, if people required D3D9 in the first place, they would be very likely to use shaders.

What do you think the "DL" in "DLSS" stands for? :)

I think he (or they) think about things like GANcraft not DLSS.

Same for "GAN" in "GANCraft", then?

I've heard multiple times the success of many games come from the gameplay mechanics, level design and story, not necessarily from the graphics? So I can see the incentive for certain future studios to maximize energy and resources on grayboxing and gameplay discovery and less on hand crafting textures and shaders thanks to generative techniques (probably in combination with capture techniques like gsplats).

This day is far of course, current GenAI research experiments are mostly useless and ugly (only the tech bros and researchers get excited about them). So the disruptors will need to invest a lot of resources on making the tech fully art directable, runtime efficient, ethical, and legal. Assuming those problems can be all cracked simultaneously, that is.

Haha yes was aiming for the dumbest possible idea I can come up with ad hoc, but of course my severely limited imagination is no match for reality.

“GANcraft essentially turns every Minecraft player into a 3D artist!” - blurb on GANcraft page. Go for it folks, now you can be artists to, get back at those evil gatekeepers and their Blenders and whatnot. Steal their thunder in the name of AI justice! Shit, I threw up a little in my mouth yet again.

As opposed to Doom in a fragment shader? ;-)

How about "live and let live" principle?

People use fragment shaders for other things too.

So you don't see how it's related?

Or should I mention all those 1FPS shaders coming from no other than iq (note: they were 1FPS at the time of release, not on the latest GPUs of course).

The techniques behind them are useful indeed, but in this particular shape and form, it is just a show-off, which is kind of similar to this stable-diffusion based Doom.

Will it be ever useful - time will tell. But why cry about it today?

Shadertoy is fun but always amuses me how bloated they are from a technical perspective, generating the same mesh definitions and textures per pixel every frame. And shadertoy effects instead of traditional geometry with light shaders applied later seems to be the norm even in big demos nowadays (I guess because it's easier and smaller to experiment with and make something cool).

Besides that, I always found the global warming alarmism around certain weird hobbies people have too cringe, considering we might be "boiling" the planet in a lot of other ways just by too many humans and industry doing all sorts of things. But people like to single it out at bitcoin and AI these days. Which I understand a bit, for example bitcoin seems like a pointless pyramid game to me without producing anything so I guess all these GPUs could be put to something better, but I am kinda biased when the global warming alarmism argument comes in for these things because people don't like them and it's associated to some tech-bros hype.

I can't speak about others, but what I point out about this or any other tech is not "global warming alarmism" - it's criticism/measure of the environmental impact.

1) I point it out because this tech does not justify the amount of its energy consumption, this tech (in this toy form we're talking about) is redundant and wasteful. For example, it's downright crazy to use all that tech to enable a person with no talent to effortlessly produce a picture that is in fact inferior to what a talented 12yo can produce with a simple pencil, don't you think so? It's the same thing with NFTs.

2) Whataboutism in this case amounts to idiocy. Because environmental impact is cumulative. So, anyone that goes saying stuff like "yeah, but email reading consumes energy too", seems to be oblivious to logic - the environmental impact of our new toy LLMs adds to the environmental impact of pre-existing tech. Just look at all the new gas turbines Musk's X put in operation to power servers for Grok image generator. Not to mention additional fresh water consumption of this facilities.

The fact that I'm critical about this tech from other angles does not discredit this important angle by any means. Please think, people.

Yeah, LLMs and climate change is serious matter. But this is off topic here.
But even if, let's estimate. Playing original Doom should consume 10W tops on modern hardware, this is of course the preferred way. Now, playing Shadertoy's Doom can consume easily around 200-300W. This stable-diffussion Doom uses single V100 TPU to run and those TPUs have similar power consumption to GPUs (ofc that doesn't count training time). But yeah, we should stop both activities ASAP.

That makes sense. AI are doing things quite the bruteforce way and just by using tons of old data, maybe there are better ways. I just don't see any way progress will stop, if there is more bruteforce technology to consume, more people will abuse it. AI or BitCoins might be irrelevant to me (I prefer for example to procedurally generate textures than use AI to get them, gives me more control and I excitement that I am engineering this myself) but it's a pandora's box now and resistance is futile. Sorry for being cynical :)

The main thing I have against LLMs is that Doom is currently obsessed with the tech side of it, and I'm finding it difficult to convince him he should put it down for a bit and make more Amiga demos. But I'm working on it.

Yeah, this is scary. Cynicism is understandable in such situation. Did we deregulate ourselves out of any possibility of control over corporations? Damage control? Possibility of stopping our extinction? To at least steer the wrecking ball away from our children's bodies if not being able to stop it altogether? If this is indeed the case, then it's all already over. The system is broken. This artificial alien entities called Corporations are Skynet. And they are already in control of humans. Thinking that some human cabal is in the drivers seat is so naive, it's beyond childish. Something infinitely more sinister, more relentless and more inhumane is in control. (gregorian choir music cue)

Oh crap, I'm pissing my pants again! ;-) Sorry, never mind me, I'm just the "crazy old man in the church", back to topic...

I see someone else also enjoys the writings of Charles Stross…

Yeah I wonder when people will finally start using shaders for something.

Gargaj: not in this shape or form..this was crucial bit..they also already use AI and GenAI for something whether you like it or not

Okay so shaders were introduced around 2001 (GF3) and shader-DOOM was 2014, roughly a year after Shadertoy opened, and about 6 GPU generations later. The utility of shaders as well as the capacity of what they can do grew over that time, as well as people's understanding of them. They remain a technology that is fairly niche in terms of understanding (and ripe for misunderstanding, but that's a different story), but they are a fundamental building blocks of realtime computer graphics today.

If you consider LLMs, they've been barely out for 2-3 years, and by the last year or so they've taken over every single news headline, with the stakeholders mostly touting how it's going to "discover all of physics", while the tech itself is incapable of counting letters and simultaneously has developed a carbon footprint big enough to rival a day's worth of airline traffic.

So here's the thing: I'm okay with the tech being "not in this shape or form" - all tech needs improving - but the comparison to shaders requires such pretzel logic it's not even laughable. Shaders were a tight, functional technology that did and does what people wanted, while in its current state LLMs are an inefficient solution searching for a problem, at best at an alpha level, that is already being bolted on to everything whether you like it or not.

The worst part is that you also know this, we know you know this, and you know that we know that you know this.

Hmm, maybe I am not clear enough. I am not talking about utility of shaders as a whole, but utility of full-screen raymarching shaders that were historically slow to render, therefore had little use outside of demoscene context. This is the "form" I was referring to.

Here the form is rendering and simulating game logic of Doom using AI, which like I said, is the most inefficient way imaginable. But it doesn't discredit IMHO AI or GenAI as a whole. Also NNs, diffusion process, transformers etc..are very well understood and you can also say tight methods. The fact that some people use it in production to ridiculous tasks that consume ridiculous amount of energy doesn't discredit fundamental research on those methods.

Back in the days Doom in a fragment shaders was the shit and got some decent spotlight, even Carmack was impressed. Today, it seems AI-Doom gets pretty big spotlight. Both equally useless *as is* for games. You don't see Doom Eternal rendering as full-screen quad raycasting per pixel, do you? How is that different?