Well, I would like AMD to offer some sort of alternative to PhysX that's comparable in terms of performance and capabilities. I was expecting Microsoft to offer such a solution, at least for Direct X, when they purchased Havok. And then, a couple of years later they announce Direct Physics, so again, I expected it then. I know a lot of games that have PhysX support use it for eye candy over gameplay, but it's also a feature that I think drives graphics into next generation territory. I can't think of a single game that's really physics heavy that doesn't use the nVidia solution, and if both Microsoft and Sony stick with AMD, that's not a solution that is going to be available on consoles. Shifty's point about it being a GPU task now is valid, except that it's only on one vendor's GPUs and that's not the one that's in the console space. I don't care if they have to include a special physics accelerator or if they just start supporting GPU acceleration via Direct Physics, or find some other way. Honestly, I'm sort of waiting for the game that has PhysX support on PC and Switch but not on PS4 or Xbox One.
Basically what I'm saying is, go play any game with PhysX support on an nVidia GPU and you'll see stuff like cloth simulation, tons of debris and smoke that looks like smoke and compare that to what we got on Xbox One and PS4. Hell, Cryostasis came out in 2009 and had ice that melted into water that would flow across surfaces and around objects while the smaller droplets would pool up and slosh around as a larger puddle. It wasn't perfect, but it's still much better than anything I've seen this gen. That's what I want.
I hear you, but i think you put your request to the wrong address. Hardware wise AMD has the same capabilities as NV. Anyone can implement all the features you want on any hardware. Unlike raytracing, there is nothing specialized to physics in NV GPUs and never was.
This can be done by game / middleware developers themselves, no need for a helping hand from AMD or Microsoft.
Personally i think physics simulation has more potential than photorealistic rendering or AR/VR, and the animation driven focus in games is the industries biggest failure. But it's not a hardware issue.
If you ask me what's the reason physics are so underutilized in games, it's two things:
Physics is hard and the usual practice 'fake it until it looks right' does not work as well as in graphics.
And second, engines like PhysX, Havok, Bullet etc. are somehow limited, which leads to the wrong assumption we can't do better.
Examples for the limitations / hacky solutions i see:
Dead ragdoll lying on a floor starts to jitter - need high sleep threshold to fix this.
Player holding objects, objects jitter on collisions - use a gravity gun forcefield effect to make the jitter look like intended.
Large mass ratio causes instability - give both the player and the vehicle a mass of 1.
Robotic simulation not robust (jitter, soft joints, weak motors or none at all, solver explosions) - use animation and scripting instead.
I say this with (dated) experience from those engines: My own, ODE, Havok, Bullet.
None of those allowed me to do the things i wanted, which is powered ragdolls and replacing character animation with simulation. (Natural Motion solves the problem by running their own solvers on top)
At a time i already gave up, i tried the Newton engine made by one man and surprise: It worked. Traditionally this engine prioritized accuracy over performance, but actually perf is good and comparable to the others. It was used in games like Penumbra, Amnesia, Soma.
(Penumbra has a big focus on physics and it's world is very believable / interactive - better than HL2 IMO. So i recommend it if you're curious. Feels similar to Cryostasis otherwise but no shooting.)
I have no experience with particles, but with cloth simulation. Those things surely are easier to do and current GPUs / APIs are well suited i would say.
