CMv5 --- AI image upscaling --- imagine this?

[markshot]

These games average a decade old.  And the graphics show it.

I think you have all seen what can be done via AI upscaling (not so much resolution, but detail, articulation, textures) ...  We are talking GPU's of the last few years whether NVidia RTX or AMD's FSR drivers.

So, imagine engine v5 adds an AI rendering post processor.  A HUGE CM battle at most is going to use 3Gb of VRAM leaving most gamers with 4Gb+ of VRAM free for post processing.  Imagine what could be done to put a brand new coat of paint on all of these games going 64Bit and AI image quality post processing.  Any of you who have high end machines and have played with open source AIs like Stable Diffusion know that photo realistic CM is possible.  In fact, there is a new SIMS-like game in development INZOI which is truly photo realistic.

Now, you may say, we want our compute doing game AI and physics ... not wasted on graphics.  Prior to 2020, this was a valid concern.  But all this is is being done with the GPU/VRAM via a fine tuned neural net for combat scene output.  These LLMs are getting smaller every day.  Microsoft has its PI series of 1/1.5/3 B.  There was open source Tiny LLAMA of 1B recently released, etc.  One of the trends we are seeing in AI is open source models is LESS IS MORE.  The foot prints are getting much smaller with fewer parameters and improved quantization (think of this as a form of compression; what it really is layer weight compression).  All of this just blew up in just 2023 alone.

NOTE:  BFC does not need a GPU Cluster to make a fine tuned LLM.  You can do this using the LORA technique which adapts matrix mathematics making such tuning totally possible with an RTX 4090 ($2000 USD) or an RTX 6000 ($10,000 USD with 48Gb VRAM).  A LORA recognizes that the entire neural net does not need to be retuned.  Just altering weights of some dimensions of one layer will give the same affect.  So, yes, building an LLM from scratch could run $50M USD, but a LORA for fine tuning we are looking at 5 figures USD.

I built last year a new PC with an RTX 4090 with an i-9 13900KS.  What does a heavy CM workload look like?  Basically, you have two Intel 6GHz Pcores handing CM back and forth as the other Pcores are capped at 5.6GHz.  (there is little point overclocking as Intel has done this in firmware already; so the only real tuning is to use software utilities which dynamically manage Pcore/Ecores affinities to get the most out of games; the only thing I touched in my BIOS/UEFI was to undervolt by 0.06v for a cooler CPU)  Most of you probably know that CM is CPU bound.  In terms of my GPU/VRAM clocks while CM is chugging through HUGE ... the 4090 is at 50W and yawning.

My point is using AI post processing to give photo realistic CM is going to give complete free graphic cards something useful to do.  Better graphics done via AI is not going to eat into physics, path finding, or AI.

I don't know if Charles and Steve read the forums.  But John as an Admin, I know you do.  AI is not just simply replacing jobs like programmers.  AI will be replacing products which fail to leverage it.  We are probably just a 2 years away from seeing a Crusader King like game which does not run with scripts, but instead is essentially an SP multi-player game where the rest of the NPCs are LLM personas.

So, just imagine our beloved CM post processed by AI capable GPUs.

Thank you!

[Redwolf]

Why would you want upscaling if you can just buy a GPU that does the target resolution natively? Sorry if I misunderstood what you are suggesting.

[markshot]

19 minutes ago, Redwolf said:

Why would you want upscaling if you can just buy a GPU that does the target resolution natively? Sorry if I misunderstood what you are suggesting.

I was just using that as an example as that is what most people are familiar with.  AI can add lot of detail to scenery without changing the resolution.

[Bulletpoint]

The better the graphics become, the more the basic flaws and shortcomings in the game will stick out. So I would prefer that they work on improving those instead of adding photo realistic visuals.

[Thewood1]

They can make a lot of progress by just using more modern and universal graphics libraries.

[SgtHatred]

That's a lot of buzzwords without any understanding behind them at all. You must be a manager somewhere.

[Dynaman216]

AI may replace programmers someday, but today is most certainly not that day.  I've tried AI code generation to help me a couple of times when internet search was coming up with wrong answers - the AI came up with exactly the same wrong answers, it just was more assertive in the incorrect answers it gave.  To put it another way, AI for programming at least, is currently nothing more than a fancy internet search algorithm.  

[Thewood1]

AI doesn't pull stuff out of thin air.  A number of my clients are working on AI-generated code and they are finding the AI algorithms are just recycling the same code available in open sources.  It can just do it a little faster.  AI can be a productivity tool.  But its a very long way off from anything more than raising the floor on knowledge of for basic programming. 

[markshot]

Folks,

I am not here for a debate.  I have seen computers from the very first degree programs and NYC having 3 IBM 360s with RJEs, punched cards, and assembly through today.  For those who are in the CS field, then you are the first to know that AI ~ 2015 and beyond is not CS.  It is an entirely a new field.  Only, the general public erroneously conflates CS and AI.  AI is modern alchemy where emergent properties has truly shown that the reductionist top down engineering thinking that many of us were trained in is but a single way to approach things.  Alchemists they may be; short on the formalism of Shannon, Turing, and Goddel long on guess work, but they are getting results.  And these results are having a huge impact.  * Mathematicians kindly call them "experimentalists".

CMBO was breaking new ground.  Taking our once BG hobby (I never was one) from the early isometric maps in SMG by Sid Meier and the early psychological states of CC by Atomic to break new ground.

CM has largely been CPU bound and single threaded.  Thus, in 2024, most of us are seeing CM hit the single threaded clock speed/RAM speed wall while other cores sit idle.  With regards to videos cards ... GPU/VRAM is largely idle.  Getting CM and many games to multi-threading load balanced is a hard problem.  But post processing using the AI abilities of graphic cards is mainly another step on the rendering pipeline which is quite doable without tearing apart the existing game engine.

How is AI fundamentally changing our world?  Productivity is sky rocketing.  There is a race to leverage it.  This is just one of many tech cycles I have lived through.  We are in the frontier gold rush phase.  The other day I counted more than 100 ventures doing RAG and TALK WITH YOUR DOCS using variants of LangChain, Semantic Embeddings, and Vector databases.  In 3-5 years, like any tech cycle, only handful of those 100+ firms will exist and be among the standardized recognized solutions.

The console/PC game industry is going to be pulled along into the same cycle.  Those with IP and a market position are ideally suited to be the survivors.  Or if they let momentum and sloth make them feel secure, then be among the shake out victims.

I like CM and the innovation that BTS/BFC brought.  I would like to see it weather the AI storm.

Well, I am going to drop off the discussion now.  Time is limited and I retired to play games and not debate on forums.  Implied here is that I might be old.  Yes, I am old.  The view that age is a handicap is not a global view among all cultures.  Some cultures actual recognize that seniors bring something to the conversation.  Tik-Tok on, folks!  I am going to play myself some CM!

Edited February 13 by markshot

[Bagpipe]

Someone's been reading too much speculative journalism. Either that or just bought shares in some AI company...
BF are hardly the market leaders in showcasing innovative computer gaming technology so I think your words are all just bluster with little real thought behind them.
Think a few deep breaths are in order and maybe some meditation or something.

[MikeyD]

Some console games cost as much as franchise superhero movies to produce. If you want BFC to spend $ 100 million developing their next game you probably wouldn't like the price tag at the end.

[A Canadian Cat]

On 2/11/2024 at 2:07 AM, markshot said:

Any of you who have high end machines and have played with open source AIs like Stable Diffusion know that photo realistic CM is possible. 

It would be super cool to see soldiers with four arms handle two MGs at once and 12 fingered soldiers will be better for throwing grenades.

That's what I currently think of when I hear AI image generation.

On a side note someone even 3D printed an add on finger you can wear so you can later claim that any video capture of you was actually a deep fake  

[LongLeftFlank]

 

On 2/12/2024 at 11:10 PM, markshot said:

Folks,

I am not here for a debate.

 

 

Oh, oh I'm sorry, this is Abuse....

Ah, don't let these grumpy old cranks put you off! 

Exploring possibility space is just fine IMHO, so long as the substantial gaps to meaningful implementation are also appreciated.

****

https://openai.com/sora

(LinkedIn commentary): If you think OpenAI Sora is a creative toy like DALLE, ... think again. Sora is a data-driven physics engine. It is a simulation of many worlds, real or fantastical. The simulator learns intricate rendering, "intuitive" physics, and long-horizon consistency, all by some denoising and gradient maths.

I won't be surprised if Sora is trained on lots of synthetic data using Unreal Engine 5. It has to be!

Let's breakdown the following video. Prompt: "Photorealistic closeup video of two pirate ships battling each other as they sail inside a cup of coffee."

https://cdn.openai.com/sora/videos/ships-in-coffee.mp4

- The simulator instantiates two exquisite 3D assets: pirate ships with different decorations. Sora has to solve text-to-3D implicitly in its latent space.

- The 3D objects are consistently animated as they sail and avoid each other's paths.
- Fluid dynamics of the coffee, even the foams that form around the ships. Fluid simulation is an entire sub-field of computer graphics, which traditionally requires very complex algorithms and equations. 
- Photorealism, almost like rendering with raytracing.
- The simulator takes into account the small size of the cup compared to oceans, and applies tilt-shift photography to give a "minuscule" vibe.
- The semantics of the scene does not exist in the real world, but the engine still implements the correct physical rules that we expect.

Next up: add more modalities and conditioning, then we have a full data-driven UE that will replace all the hand-engineered graphics pipelines.

Edited February 16 by LongLeftFlank

[herr_oberst]

Yeah... but when the GPU visual graphics 'effects' render a cannon fire hitting given 'reasonable' results, how does that translate from the graphical rendering of [position a, fires at b] translate to (I'm sure greatly simplified)

[firing position a, mass of shell b, muzzle velocity c, shot fall function(b, c, d: distance to target, e: time to target, f: motion of target), g: position of d when shot arrives], determine impact point and effects on d work out? Non-trivial. Who (cpu or gpu) does the math? Does cpu 'trust' gpu, and how do you code for differential gpu capabilities? 

That's why I pay BFC to figure it out. They know better than I what the engine can do, and what moving from 'it can do X" to "it can do Y" will cost and the expected revenue stream. I hope that the consumer version will see some trickle down economics for improvements, but the consumer base likely has very different 'min architecture/hardware reqs' than their more 'advanced' clients.

<opinion>All in all, its still the one game I go to for entertainment and 'deep thought' gaming. From good scenario builders, shallow thought will get you punished... Graphics are dessert, plans, tactics, and execution are the meal.</opinion> And I realize it as such. Just my $0.02.

[chuckdyke]

3 hours ago, herr_oberst said:

, how does that translate from the graphical rendering of [position a, fires at b] translate to (I'm sure greatly simplified)

Trajectories are modelled visual effects I think are not.

[LongLeftFlank]

13 hours ago, herr_oberst said:

Yeah... but when the GPU visual graphics 'effects' render a cannon fire hitting given 'reasonable' results, how does that translate from the graphical rendering of [position a, fires at b] translate to (I'm sure greatly simplified)

[firing position a, mass of shell b, muzzle velocity c, shot fall function(b, c, d: distance to target, e: time to target, f: motion of target), g: position of d when shot arrives], determine impact point and effects on d work out? Non-trivial. Who (cpu or gpu) does the math? Does cpu 'trust' gpu, and how do you code for differential gpu capabilities? 

That's why I pay BFC to figure it out. They know better than I what the engine can do, and what moving from 'it can do X" to "it can do Y" will cost and the expected revenue stream. I hope that the consumer version will see some trickle down economics for improvements, but the consumer base likely has very different 'min architecture/hardware reqs' than their more 'advanced' clients.

<opinion>All in all, its still the one game I go to for entertainment and 'deep thought' gaming. From good scenario builders, shallow thought will get you punished... Graphics are dessert, plans, tactics, and execution are the meal.</opinion> And I realize it as such. Just my $0.02.

Sure, and I come to praise the Brainjar not to bury him!

While I am not, and never have, coded anything more complex than an Excel VBA macro, I am in general agreement with what I think OP @markshot was asserting, to wit:

...that the new generation of AI tools are already making it easier and cheaper for small devs like BFC to layer highly pleasing and 'up to date' looking gaming graphics atop their core proprietary physics and tactical decision engines.  So they can stay focused on refining that engine and the player interfaces ('Follow Me' commands, etc.) while 'outsourcing' most of the crowd-pleasing visual effects.

Again, all easier said than done, but our dear sponsors have also got to move with the times, elsewise get bought out and folded into a larger shop that won't allow them full creative control.