If you read nothing else, just take a look at the physics engine examples I linked below (particularly the shatter engine). They will explain more succinctly than I can what the future of physics engines will be.
Tanakh said:
For AI... not really. Video game AI is not limited ATM by solely by the processing power, but rather due the paradigms it follows. There is no game that uses anything done by AI in the last 25 years AFAIK, wouldn't it be possible for games like Skyrim or LoL (bots)? Of course it would, fuck I could do it myself with current tech but that's not the trend. For physics? I guess... but tbh i am quite content with current physics in the PC, battlefield 4 seems as good as I need in that regard, even the BF3 for the most part tbh. About the price? Not bad actually, nice hardware for a decent price, I might buy one at some point but calling it "that ambitious" is a gross exaggeration imo.
AI is currently used in games with bots like COD Black Ops II for example and more generally in how non-player characters and enemies behave in any situation. The way enemies respond to human interaction is basically what I mean by AI. Artifical intelligence isn't necessarily something grand like we see in movies, it's just how an artificial entity sizes up the situation given an evironment that can be every bit as detailed as real life. Are they aware of where the character is standing and where they need to be in relation? If they can get it to react the way various humans might and even create different digital "personalities" that would be a huge improvement. Currently, when I turn the corner and see a bot magically pointing a gun directly where I wasn't a few seconds ago but am now, that makes things incredibly apparent. It means that they're not responding, they're aware of everything in a way we can't be while making decisions on what they can respond to and what they can't. That can change.
Presently, the number of bots you add in a multiplayer map in COD Black Ops II decides whether or not the console will stutter and crash at some point. That's because even mild AI takes up some serious resources. It has to process interactions in real time. There's a reason why we have sooo few games that have successful bots. Ever let a soldier drive the warthog in Halo? Might as well just go ahead and jump off the edge of the map.
Physics covers the most, by far. For you to say that you are content with physics indicates to me that you may not be aware of what a difference they can make. How the character models move in relation to the environment around them. How does a bullet impact an arm or wall. Take a look at some advanced physics models and see what differences it can make:
Shatter physics. [http://www.youtube.com/watch?v=FIPu9_OGFgc] Note that I debated whether or not this was real until I saw it having trouble rendering the appearane of glass at the 4 minute mark. Started seeing more indications later.
Wireframes, rope, chains, spaghetti modeling. [http://www.youtube.com/watch?v=i1YIafaqF7A] Ultra realistic, same people behind the shatter physics.
Particle and vector physics (temperature, wind, smoke physics) [http://www.youtube.com/watch?v=KQMZ2mE-8HI]
Water physics demo [http://www.youtube.com/watch?v=JcgkAMr9r5o]
This is specifically a car impact physics engine. [http://www.youtube.com/watch?v=A4C0zM2vnsQ] The 45 second mark is when stuff happens. This is just an early protoype of the stuff they're starting to do. There's a lot more that can be tweaked and added to make some REALLY interesting stuff.
If you like the videos above. I recommend perusing Phymec's youtube page. Lots more incredible stuff: http://www.youtube.com/user/Phymec
Needless to say, there is a TON of room to grow in the physics department. Every time you see a shirt tail disappear into a character model's body or anything like that, it's bad physics that has no understanding of its environment and it does break realism.
RAM has gone from $0.185 per MB to $0.0054, that is almost wooping 3500% cheaper! The change from 0.5GB to 8GB is much smaller, and in both cases the PS is using standard gaming quality components, which is good but again nothing too exciting.
You made a mistake here. You did a price difference rather than a power difference like you did for the video cards. There is no power difference in what pc's can generally handle that would be 3,500 times more powerful than what we had in 2005-2006. If that was the case, the PS4 would have to have 1,750 GBs of RAM to match your spread when coming from .5GB of RAM.
The ps4's RAM is actually even a little more than 16 times the ps3's. First off, you get the 16 times more automatically just numerically because 8GB is 16 times as large as .5GB. Then you have the fact that the ps3 was arbitrarily divided into two 256MB sections, not unified. And that it wasn't .5GB of DDR5 which the 8GBs is. I don't know what number that ends up at afterwards but it's a very respectable number comparatively. And the thing is, CPU's offload to the RAM so this is a more important number than CPU.
For the CPU... kinda harder to measure. But if you take GFLOPS per dollar you will see that again, it has been increased by way more than 10 times what it was in the PS3 release. You will also see that it seems the PS4 is the cheapest FLOPS in the yard ( http://en.wikipedia.org/wiki/FLOPS ).
With the availability of cheap RAM, CPUs are no longer as important as they once were. They serve as something that manages the offloading of processing of various applications to the RAM and GPU. What you're seeing here isn't necessarily a failure at all. It's just a change in the way computing used to be. If the CPU is working really hard nowadays, you're doing something wrong. So what you want is a competent CPU with multiple cores and hyperthreading just in case someone ever wants to use that. That's what this has.
Forgive me if I'm missing something though, but what does the cost of the flops have anything to do with it? That's just saying that for $400, the ps4 offers the most number of flops for that price. That's actually a good thing that speaks to the value of what you're getting. For example, the option immediately above it at .75 per GFLOP was $3,000 to build 4 TFLOPs in 2012. $400 for 1.84 TFlops is great. 1/3 the cost per GFLOP in only one year. I'm afraid you'll need to clarify your point here. I must be missing something. $400 for just under half the TFLOPs $3,000 could get you a year ago? Sounds great.
My conclusion is that it will be an extremely good deal for entry level gaming with a robust library, definetly the best deal on consoles by far regarding hardware and probably better than the PC you would get for $400 when it launches, not a bleeding edge amazing gaming machine by any means, but it fits a consumer.
Regardless of my disagreement with the points in particular, I fully agree with your conclusion. It's a great deal for $400 but by no means a powerhouse. The gaurantee that games will run on it as long as it is being sold also adds a lot more to the dollar value as well. The fact that the hardware is standardized will also mean that dollar for dollar, developers will get far more out of every component than they would be able to get out of comparable pcs whose components were cobbled together. This isn't going to be the last step in the ladder but we will always need consoles or at least a standard by which developers can plan for. Everyone wins with consoles.