Hacker News Reader

Comment by avenoir | Hacker News Reader

I was looking at the animation referenced in the article of water flowing over a rocky terrain and couldn't help but notice almost life-like and pixel-perfect precision of collision detection between water particles and the rocky surface. Yet, in modern gaming, while it's improved a great deal, quite often you'll see surfaces "fusing" together before a collision is detected. I think there is a more technical term for this, but it's not coming to mind. Is this just a matter of usability vs performance?

View on HN

I would imagine its a performance issue. For Moana they can render the water on a server farm with has much computing power as they need. Where a console or personal computer is quite limited in contrast.

Further to that point they can take as long as they like to render a single frame.

The movie is using a whole compute cluster to do a physics simulation when rendering the movie. Your computer has 1 CPU and 1 GPU.

Your computer is just very limited in how much attention to detail it can have, because it has wildly less processing power and has to get its job done in real time.

By contrast, Disney is using a server farm to spend 3 seconds analyzing the mechanics of a tuft of hair wiggling in the breeze for 1 second.

Has anyone tried to make a real game in the same way?

Ie a game which was hosted on a remote desktop that had direct access to a GPU server farm on a very fast network.

Not a field I work in but I guess it would never work. You'll have the network latency to contend with. Or people would need to self host a server farm, which probably makes your market pretty small :-)

I wouldn't say never. The golden rule of computer games code is that you get to cheat a lot.

https://www.playstation.com/en-us/explore/playstationnow/

It certainly could happen - http://www.nvidia.com/object/cloud-gaming.html - how much will someone pay per hour to play one game? It feels like a timewarp to pre-AOL when online games cost more per hour than minimum wage.

http://www.nvidia.com/object/cloud-gaming.html

This was initially attempted by G-Cluster in 2000, Crytek tried it as well, popularized by OnLive (who started working on it in 2003) and blew up in 2010.

The general term for it is cloud gaming https://en.wikipedia.org/wiki/Cloud_gaming

The answer is no. People are bringing up OnLive and Nvidia but they are just running the same game you run locally on remote hardware.

What you are asking is basically has anyone run a game on a supercomputer/cluster so that the graphics are way better than what could be done locally. The answer is no, nobody has. That would be pretty neat though. It would be super expensive to make with no way to recoup the cost.

Here is an example that is not what you are looking for but is somewhat related. It's a game that is played by rendering high quality screenshots of the game world on a server. I enjoyed it a lot.

https://extrasolar.com/

https://lg.io/2016/10/12/cloudy-gamer-playing-overwatch-on-a...

Additionally, multiplayer games often have to run a low-geometry simulation server side to reduce cheating and account for prediction/latency sync issues.

While certainly not directly linked to poor physics simulations of water in computer games I would guess it has played a big role in holding back advancements in this area because how much locally-only non-game mechanic influencing visuals could cause contrast with need to be duplicated on the server physics.

The other thing to remember is that Disney are completely at liberty to reject and re-render scenes when something doesn't look perfect. Arguably, this even applies to examples chosen for the demo video.

To put this in perspective, 60fps == ~16.7ms per frame.

That's 16.7ms to run your WHOLE game engine, you get some fraction of that left over to do rendering.

The quality of game renderings will lag behind movie renderings for a long time. At 60fps there is about 16.7ms available to render each frame.

For a movie[1]:

> There's a principle I call The Law of Constancy of Pain. Back when we made The Adventures of André & Wally B. in 1983 a typical frame took from several minutes to a few hours to compute. Now computers are literally millions of times more powerful and a typical final render takes from a few minutes to a few hours to compute, averaging somewhere between 2 and 4 hours. Render times have remained flat for over 30 years because that's just how long people are willing to wait for them.

So movie artists are willing to have each frame of a movie take 2-4 hours to render vs 16.7 milliseconds.

[1]: https://www.quora.com/How-long-does-it-take-to-render-a-Pixa...

The thing I noticed was how there seems to be some sort of hydrophobic coating on the rocks. They don't get wet like real rocks. The lava demo is similar.

Reply on news.ycombinator.com