Nvidia PhysX 5.0 | Hacker News Reader

Nvidia PhysX 5.0

github.com

292 pointsvectorrain109 comments

For what it's worth, PhysX 4.0 (2018) was also Open Source (3-clause BSD license).

Previous discussion here: https://news.ycombinator.com/item?id=18589494

Ok, we've edited that bit out now. Thanks!

(Submitted title was "Nvidia PhysX 5.0 is now open source")

Note this isn't really fully open source as the CUDA kernels for PhysX aren't distributed and the HLSL code for Flow isn't included either (only Vulkan and DX12 bytecode is included, no sources). Happy to be wrong if someone finds the code for these :)

What are some games that have impressive physics nowadays? I remember being very impressed with Half-Life 2 back in the day, seemed like a leap. All spectacular looking games with crude physics nowadays remind me of those hi-res ports of old console titles, in an ajar way.

apart from games going from 2k to 4k to 8k, there hasn't been any noticeable improvement in realism in the past 10 years. ray tracing if enabled looks nice but that's pretty much the height of it, characters look just as janky as they did 10 years ago.

I feel the primary reason are the consoles, game devs can't push the boundaries as most consoles are around 5-6 years behind gaming PCs.

Have you played The Last of Us part 2? The amount of details all over the world, smooth animations that blend midway into another, etc, really is a step forward imo.

I think your comment is wrong and perhaps speaks to your own visual senses than any factual effect.

There’s been so many advancements in the last ten years outside of raytracing. Better character motion, better character AI, spatial audio, audio materials, speech matching just off the top of my head.

Every single thing about games has gotten noticeably better for realism.

One only needs to watch a Digital Foundry video or a GDC talk to see the big uplifts.

10 years is a long time too and spans all the way back to the PlayStation 3 and Xbox 360 era. So your comment is nonsensical at best

character motion being the same old animations, just done/interpolated better

the only advancement I know of which got close to be implemented in games is https://github.com/sebastianstarke/AI4Animation (I think Sebastian worked with EA or some other big company at some point) - still haven't played anything using this though

What a surface level take.

Firstly, we’ve been doing the “same old animations” for a very long time, and in most areas of content.

Secondly, they aren’t they same old animations, technologies to produce them are becoming higher precision, more efficient, and better in many ways.

Continuing on, the amount of work being poured into dynamic animation, IK, and the like is significant.

Various locomotion systems, including AI4Animation that you linked are becoming significant contributions. Though crazy you single that out as an exception, because it is also using the “same old animations” just a large unstructured set of them. But this work is a direct continuation of motion matching, which also works on large datasets of unstructured animations, and *has* shipped in quite a few titles, and is a very significant jump in how animation is done today.

And that’s just animation.

https://www.gdcvault.com/play/1023280/Motion-Matching-and-Th...

Your opinion here is highly ignorant.

There’s been tons of advancements on higher fidelity motion, better interaction , pose blending, constraint targeting, secondary motion, pose based deformation etc..

10 years is two console generations ago, 2012 was still ps3/xb360 era. If you don't notice any difference, then that is more explained by your lack of paying attention than lack of improvement.

Cyberpunk 2077 at least pushed the rendering part of the engine into next-gen territory: Volumetric lightning and fog almost everywhere, environment reflection on even the smallest water surface and monochromatic light actually turns many surfaces into mirrors. All that in an outdoors setting with a ton of assets and an almost impracticle amount of verticality (no naturally empty half of the Screen). Before that release, most game devs would have told you that this is still impossible on todays hardware.

But that engine also shows that you are not wrong with your remark about consoles holding progress back: The game barely ran on the ps4, and even on the ps5 you won't get stellar framerates... You need a really nice PC to run the thing smoothely...

It runs really, really smoothly on my Xbox Series X. I'd arguably say even smoother than my OC'd 10900K + RTX 3090 PC.

The pursuit of realism hit diminishing returns years ago; the manpower needed to make a game hyper-realistic isn't justified by the number of additional sales it will produce. Instead of becoming more realistic, most games have instead focused on having more bright neon particles swirling around on screen. That's much cheaper and much more effective than chasing further realism.

This will likely change in the near future as the industry invents ways to replace human modelers, texture artists, etc, with automated "AI" tools.

You are right, but I think it goes back and forth. I can recall seeing a game I was working on (Homefront 2?) where it looked surprisingly photo real - at least as a prototype. But the physics took you out of the illusion as soon as anything moved.

Nowadays I see developers doing amazing things with shaders, like you said. And stylized models are easier to make, but also less in the uncanny valley.

Realism can be cheaper when you don't care about optimizing, or AI can optimize for you. When you can scan in objects and it is workable in a game engine that's cheaper than designing them. We're there with mo-cap vs character animation.

It's not quite as easy as slapping on a good physics engine and call it a day.

A good integration with a decent physics engine feels better in-game than a superb physics engine with a bad integration.

When programming physics there's many ways of getting the desired behavior from the physics system, and as with all problem solving some solutions are better/more stable than others.

I don't doubt there are compromises being made. I guess I'm just a bit surprised that immersion via realistic physics isn't in higher (and competing) demand. It also seems to me there could be a lot of interesting gameplay mechanics to explore there.

There just isn't enough talent to make those games. People who are both good at programming and physics are very rare, and without them you are unlikely to produce programs that can use the physics system in reliable enough ways.

The Noita game features huge 2D levels where gasses and liquids interact with magic spells. Teardown has completely destructible voxel levels. BeamNG driving simulation implements soft bodies, various experiments with its physics are popular on youtube. I don't think any of these use PhysX.

My only problem with Noita is that the back half of the game basically disregards most of the physics and chemistry systems. It’s more based around exploiting the build-a-wand system so you can e.g. generate infinite black hole chainsaw spells that let you blast through rock at a million miles an hour.

It is the only game I am aware of that I would say has a true “chemistry engine” along side its physics engine. (Though props to designers of Zelda BotW for giving objects internal properties besides just velocity & mass that facilitate reactions and interactions. Falling sand games that Noita takes a lot of inspiration also often have “chemistry” interactions but generally don’t incorporate physics bodies and are more of a simulation toy box than a game).

It's the great unanswered question: what works in game design?

It turns out wand building and extensive end-game/meta progression is far more engaging than most other things. IIRC there was more chemistry early on (I think there was/still is cooking?) and more rogue-like elements like satiation, but wand-building combined with the world physics turned out to be such immense fun, that it eclipsed everything else.

A Kerbal Space Program with an on ramp to the world's observatory catalogs and focus on learning instruments and physics, theoretical and applied, would be awesome from Nvidia.

https://en.wikipedia.org/wiki/H-alpha

The reason is money - it costs money (artists) to make destructible objects and art down for every conceivable way it can be broken/destroyed/manipulated.

I hope/pray for a Battlefield 7 game that actually takes this seriously, but EA has run Battlefield into the ground.

Some ex-Dice guys are at Embark Studios and their next game seems to be bringing back destructible environments.

Trailer: https://www.youtube.com/watch?v=YYYiP2cKHKs

https://www.embark-studios.com/

All I could think when watching that trailer is "I'm getting too old for this".

Also, relevant to this story, Embark does a lot of open source development in Rust, including physx-rs. The Finals is using Unreal Engine though.

Very cool, thanks for sharing

Bad Company 2 came close to this and for me was probably the most fun and thrilling Battlefield game ever.

Maybe Teardown would be interesting for you.

I followed that project and it's part of the reason I'm asking. If one guy can achieve that...

The CPU on xbox one/ps4 is very weak. Developers have to sacrifice physics.

The whole point of PhysX is to run it on the GPU.

That's not true at all. PhysX comes with a CPU solver by default, and using the GPU version comes with significant restrictions in customising the behaviour. Unreal Engine games (when they used to use PhysX) were all CPU based, for example.

Boneworks has very impressive physics and interactons, it is even advertised as "Experimental Physics VR Adventure."

Designers figured out that it's hard to make physics-based gameplay that's fun and interesting and doesn't break the game in various ways. Physics-driven games in the HL2 style are then almost their own genre - it has to be a conscious design choice - so most games don't try to make complex physics core to gameplay, they just use it as window-dressing. There's a lot of impressive window-dressing these days! But I don't think that's what you're talking about

Zelda Breath of the Wild is a recent example where the designers went all-in on physics-based gameplay and did an amazingly impressive job with it. But the hard part wasn't the technology, it was everything else.

Also realistic physics is very hard on multiplayer. The more detailed the simulation the more data you have to keep in sync.

These are practically ancient in internet time but they are good examples of physics-based gameplay.

https://jet.ro/dismount/

Seeing that just dredged up some ancient memories

Teardown - hands down one of the most innovative games in the last several years.

It uses voxels and the game premise is destroying the environment to create an escape route once you trigger the alarm (by stealing required objects).

https://store.steampowered.com/app/1167630/Teardown/

Looking at a gameplay video on YT (https://www.youtube.com/watch?v=P--c8rjcSDs), I didn't come away with the same impression as you described, or at least it doesn't really answer the parent's question. Sure, it is "fully destructible" but everything kinda breaks the same way because it's all voxels, and things like driving a seemingly indestructible truck through multiple walls doesn't feel particularly realistic.

Someone else pointed to another game, The Finals, which feels waaay more realistic: https://www.youtube.com/watch?v=rGRoborkkw4

I just hope it isn't another Battle Royale because that I find that genre incredibly underwhelming vs. team-and-round-based gameplay like CS:GO or offline campaigns

I wonder if you'll have a different feeling after seeing a trailer instead of gameplay of the first mission ;)

https://www.youtube.com/watch?v=ttwBelIlLv8

I'm not aware of any other real-scale voxel game out there. This one has water physics (things float), cars are destructible (you saw a construction vehicle that can take a lot more damage before noticeable deterioration), and more. Cheers!

gameplayvideos are pretty much always better than a trailer

The physics of Teardown may not be realistic, but that's not what was asked for - and the physics of Teardown are DEFINITELY impressive.

Getting it to run performant with everything being destructible is impressive yes, but at the same time the physics seem super simplistic. There doesn't seem to be any kind of deformation or elasticity nor lever action. All I see in the videos is just blocks breaking off of paper houses.

Uhhh, Physics are indeed a major part of Teardown. the breaking of the Voxels themselves may not be individually impressive, but it's the nearly entirely destructible Environment with per Voxel calculated Materials that IS impressive. Voxels can be of a wide range of Materials, giving them, well, unique properties. that allows things to be strong, weak, and all sorts of other properties. Driving a Truck through a Building doesn't leave it unscathed, actually. Et Cetera.

They are impressive on a technical level, especially given how well everything runs. However, I think that the other was pointing out that the game feel/presentation itself was a bit lacking - not having much momentum to it, or just the destruction itself not feeling particularly eye catching (even if the gameplay is fun, due to you needing to act within a time limit).

I'm inclined to somewhat agree, because you can still have situations where buildings can be hanging on due to a single voxel and will refuse to fall down, and when they do there's no sense of weight, instead they just kind of plop down. Very much the same how the cars and such also feel awkward.

It doesn't detract from the gameplay much, and it's not like that makes the game bad, but personally I think that the Red Faction Guerilla game felt a bit better. Of course, it was geared more towards presentation and had a large studio behind it, rather than being very technically accurate, so the goals are a bit different than those of Teardown (interesting setpieces vs procedural destruction).

Here's a few random YouTube videos:

https://www.youtube.com/watch?v=n16pZxHBo4o

https://www.youtube.com/watch?v=eXmKlVZmvRg

It is pretty awesome to see projects like either game, though! Even engines like VOXLAP were interesting: http://advsys.net/ken/voxlap.htm That's actually more or less what powered the old Ace of Spades game, which is now Open Spades: https://openspades.yvt.jp/ (not as focused on destruction, but rather having large maps with lots of voxels, a fun game)

Beamng is really impressive physics wise. Lots of mods available and endless gameplay possibilities.

Embark Studios are making a game called "The Finals" which leans very heavily on destructionn. The trailer has some really impressive building collapses and other loud looking events: https://www.youtube.com/watch?v=X84w4DWEEes

Embark was founded by a lot of the old school DICE guys. Really excited for what they bring to the table.

Also iirc they are big on Rust and are supporting a lot of grassroots projects in the Rust gamedev space

The destruction in the trailers reminds me a lot of Bad Company 2 when it was possible to level the battlefield. Later games dumbed down the destruction or had scripted destruction sequences.

They use PhysX with rust bindings for this. https://github.com/EmbarkStudios/physx-rs

The Finals is an Unreal based game, not Rust, you can see the Unreal logo on their homepage: https://www.reachthefinals.com/

They have a Rust game: https://medium.com/embarkstudios/embarks-creative-playground...

Which maybe be using their PhysX bindings but it's not The Finals

https://www.beamng.com/game/ - possibly the best driving / soft body physics out there..

Is there much written about their sim approach?

Answered my own question: https://wiki.beamng.com/JBeam_Physics_Theory.html

I'm very impressed with Trine [0] and I noticed it has the Nvidia PhysX logo on startup. The entire game is based on physics puzzles and the graphics are gorgeous.

[0] https://www.nintendo.com/store/products/trine-enchanted-edit...

Trine is a 13 year old game - released closer to HL2 than today.

Yes, I see the series has been well-received. The version I picked up is from 2018, called Trine Enchanted Edition.

Instruments of Destruction, still in Early Access: https://store.steampowered.com/app/1428100/Instruments_of_De...

BeamNG, the most physics intensive driving game ever made.

https://youtu.be/Lqx2KKWI8aM

iRacing

Death Stranding. Considering you traverse the world mostly on foot and packed up they nailed the terrain and how movement with heavy package works (imagine sherpas and mountaineers).

Maybe not state of the art but I was very impressed by Assassin's creed origins.

Outer Wilds has some impressive physics in that the whole game takes place in a miniature solar system where the planets orbit the sun, have their own gravity, day/night cycles, and so on. There are various weather phenomena, such as one planet that has floating islands tossed into space by storms before crashing back down, and a pair of planets named the "hourglass twins" that transfer sand between each other. You can fly out in your ship and watch these things happen at a macro level, or get right into it (like walking around one of those islands as it's being tossed into space, or getting yourself crushed by rising sand levels). It's really impressive, and I'm probably not doing it justice with this description. Worth checking out.

Boneworks is a VR game where almost everything is physics based, including movements of all characters and your own avatar via a skeleton system. It has a different kind of feel to it and is quite innovative. It's an interesting approach to solving problems caused by having physics based objects interect with the "unstoppable force" of traditional player movement and the "immovable objects" of the environment.

BeamNG will blow your mind

> What are some games that have impressive physics nowadays?

Tetris?

Probably should remove the 'now open source' part of the headline.

Nvidia's own marketing page about the 5.0 release.

https://developer.nvidia.com/blog/open-source-simulation-exp...

Right, that page says "All CPU source code is available under the simple BSD3 open source license, and NVIDIA GPU binaries are included at no cost." So only the CPU code is now Open Source.

Also, if you are building a game, wouldn't you want it to work on AMD GPU's too?

I seem to get unpredictable, if infrequent, hard crashing when playing the PhyX-enabled game Control unless I set the Nvidia control panel option to run the physics on the CPU. System event viewer pointed to a failed assertion in the Nvidia driver. I'm not 100% sure this has fixed the problem but I haven't seen a crash since the change.

Since gfx benchmarking is stable and MemTest86 never found anything, the only other culprit would be power transients. I'm using a relatively modest 200 watt RTX 3060 ti so I hope it wouldn't be that.

Just wanted to bring this up because I'm not the only one that has experienced this, it was a recommendation from a Steam discussion. And by hard system crash I mean full system reboot.

For the younger crowd, back in the day before Nvidia bought Ageia, you'd have to buy dedicated PhysX accelerator cards and stick them in your PC in order to use this tech in the few available games that supported it. Wild stuff.

After that, PhysX API was accelerated via CUDA and the dedicated PhysX ASICs were discontinued.

https://www.techpowerup.com/review/bfg-ageia-physx-card/

Interesting. What was the target group of these cards? Gamers or engineers and scientists?

It started out initially for academic usecases till the costs came down and it was then pushed for games.

Mostly gamers, here's a review from Arkham Asylum which showed how offloading PhysX to a GT 220 paired with a GTX 260 would help a fair bit with minimum and average FPS - https://www.tomshardware.com/reviews/batman-arkham-asylum,24...

This is from later (2012) but shows some of the effects that PhysX enables in games. The "apex turbulence" in the second half of the video is pretty cool.

https://www.youtube.com/watch?v=n5qhaEghJ74

Even after the acquisition, people still used dedicated PhysX cards before multi-monitor screen stitching made SLI graphics visual based. Prior, you’d have a card in SLI to handle the physics for PhysX and the other for the visual rendering.

This is great news. Bullet and others could benefit from this as well.

We used the PhysX engine in a simulated environment for robotics development back in that timeframe. I don't think more than one or two people on the team had the accelerator cards, but it worked surprisingly well just on a vanilla hardware

Dont forget to mention it was pure scam, PhysX PCI "accelerator" actually ran those supported games slower, sometimes Two times slower, than software.

https://www.anandtech.com/show/2001/4

You miss the important bits:

> As we said before, installing the hardware automatically enables higher quality physics. We can't get a good idea of how much better the PhysX hardware would perform than the CPU, but we can see a couple facts very clearly.

So what you call a scam is down to implementation details. In a true like for like scenario the PPU would usually outperform the CPU implementation as long as your sync boundaries were clean to get stuff back from it, and pipe the transforms back to the GPU.

I wish people like yourself wouldn’t reach for incendiary language by default. It ruins any and all nuance in discussion.

A more nuanced description is that in most cases, it was slower, and it took immense developer effort to even try to make it faster. And the cases that were accelerated were generally things that consumers didn’t care much about, e.g. fluid dynamics.

The software is the clear winner of that. The team was fantastic, and one of them (John Ratcliff) helped kickstart my career. So it isn’t entirely off base to say that the PPU was… exaggerated.

Its a game, the important bit is playable framerate. There arent any nuances when you are hitting drops to 17 fps.

>down to implementation details

Game is running 2 times slower. Nvidia favorite kind of technology. Tessellation, Gameworks, Hairworks, RTX, all run at lest 2x slower when enabled unless you buy Nvidia flagship hardware.

Did you know Nvidia physx library shipped compiled for FPU? At the time SSE2 was available for 10 years and SSE3 for 5 https://arstechnica.com/gaming/2010/07/did-nvidia-cripple-it... https://www.realworldtech.com/physx87/3/ Nvidia PR person tried claiming Physx actual computation is not a bottleneck, so they didnt bother optimizing :D If its not a bottleneck then why run it on GPU?

I'm happy to see this! Hopefully Nvidia starts open sourcing a lot more stuff.

PhysX has been source available since 2015 (3 clause BSD since 2018) , so I wouldn’t take this as an indicator in that regard.

right. what do you think about nvidia's recently released gpu drivers ?

I think it’s good , but doesn’t change too much unless you’re a kernel implementer.

They’ve been busy moving the vast amount of the bits into the GPU firmware (not uncommon, this is how Apple and some others do it too).

I think the FOSS crowd made a bigger deal of it than it was because it appealed to their sensibilities.

The best new OSS-ish stuff from NVidia is their research, and things backing that research. They’ve released a lot of their nerf tech in the wild and Warp (a differential Python to CUDA transpiler) which are very cool.

Not the person you're replying to, but those drivers are essentially just the communication bridge between the kernel and the "real" driver, which is mostly in the closed-source firmware blob. They're also woefully incomplete.

And only on PC and Mobile. Not on console.

Consoles and open source do not mix. NDAs prevent code using their APIs from being released.

That said, a lot of open source projects have console ports, they just can’t be shared publicly.

Steam Deck is in disagreement.

Steam Deck isn’t a console in any traditional sense unless we’re just going to start butchering decades of colloquial nomenclature . It’s a portable PC running desktop software from top down. Otherwise you might as well call my laptop running Steam Big Picture a console too.

When the other people are talking about consoles, any sensible person knows they mean things like the PlayStation, Xbox or Switch.

Playstation is a PC with extra lockdown. So is Xbox. The Switch is similar hardware to Nvidia Shield, which is just another general computer (it can run Nvidia's version of Ubuntu).

So the differentiator these days seems to be running custom software.

Which Steam Deck is also running.

The remaining difference is "console = extra lockdown". That might be a good definition in general, but doesn't make sense in the context of the original post

> And only on PC and Mobile. Not on console.

Another definition of "console" might be "specialized and/or custom hardware designed mostly for playing games" without regards to the software.

That definition also captures the Steam Deck. And I think it's fine. It exposes that "console" being separate from "general purpose computer" is either an anachronism or a marketing ploy.

So the key distinguishing attribute that makes a "game console" a "game console" is its walled-garden nature? Nothing else?

And not the fact that it's, you know, a "console" that can be used to play "games"?

> So the key distinguishing attribute that makes a "game console" a "game console" is its walled-garden nature? Nothing else?

Practically? Yes, IMO. You can say it uses specialized hardware manufactured on a massive scale, with each gen being a distinct set of hardware with slight variations, but then you're describing an Apple M1 Macbook Air or Microsoft Surface.

People have and will always gain root access, but the OEM doesn't typically like this, and goes out of their way to prevent it. There may be APIs the OEM leaves open to allow the creation of, for example, XMBC, but an XBox is hardly an open, general-purpose, computing platform.

If you want to call it a console that can play games, then my custom-build Linux computer console fits that definition. Hell, I wouldn't have to even leave the computer console TUI to play Dwarf Fortress.

I've worked on many console projects that use many open source libraries and some where the part that can't be shared publicly was still source available if you were a registered developer. Not disagreeing with you just sharing some info for others.

Indeed, nothing has changed. This is just a revision. PhysX 4 source was available here. https://github.com/NVIDIAGameWorks/PhysX/tree/4.1/physx/sour...

Then what about this? https://github.com/NVIDIAGameWorks/PhysX/tree/4.1/physx

That looks like 4.1

They don't follow semver. v4 is basically a completely separate product from v5.

Not to be THAT guy, but Physx is basically reaching EOL. I'm sure a few mid sized studios who want to roll their own engine can still take advantage of the core, but with UE5 completely replacing Physx with Chaos and the "ballast" around the core (Blast/Flex...) being as useless as ever, I don't see the Physx adoption increasing.

I believe NVidia develops PhysX mostly for robotic/metaverse simulation, not for games.

See positioning of NVidia Omniverse for example.

I remember back in 2000s when Cellfactor demo came out on youtube.

https://www.youtube.com/watch?v=QzIHI7y4ZG4

Which needed that separate physx card at the time. Wonder what ever happened to it.

Now they should reach for the stars and update the README with a project description.

I always deselect this when I install the nVidia driver in Windows. I play none of the few games that are supposedly enhanced by it.

I can only think of Planetside 2 from back a ways, but they removed it for the PS4 release. Cool tech, though, I enjoyed it.

You probably do, as unreal engine and unity both use physx. Just not the old system level libraries that are used for old games.

ok, cool, didn't know that

Incidentally UE with 4.26+ (and more so 5, which is almost a complete cutover) switches from PhysX to their in-house solution of Chaos.

A lot of it has to do with determinism over the network IIUC.

4.26 and 4.27 still use physx by default, and their chaos version is basically unusable[1] (because of bugs and performance). There is a 4.27-chaos branch which is more updated (maybe the same version as Ue5 chaos?) but I don't know how usable it is.

But yes the future for UE is chaos.

[1] to be fair fortnite used it, but clearly they optimized and fixed only the subset of features that fortnite uses.

Reply on news.ycombinator.com