Hacker News Reader

Comment by iandanforth | Hacker News Reader

The Stadia trial convinced me that streaming is a near-term viable option. Latencies were not an issue. Bandwidth was, but when I got to play on a 50Mbps+ connection any issues I had encountered disappeared.

Moreover I was able to play a AAA title on my 2011 Macbook Air and that was fantastic.

View on HN

What kind of games were played? Whether or not latency is an issue is hugely dependent on the title. A game like Sekiro: Shadows Die Twice where the bulk of the game play revolves around precision timing is going to be killed by any significant amount of latency. Unless it does something smart like trying to compensate for the latency on the client end (e.g. make a block succeed even if it was late, if it suspects that the player blocked at the right time from their perspective when viewing the delayed stream) I can't see how a game like that would be playable on Stadia.

I believe they're talking about the Assassin's Creed: Odyssey trial.

Yep!

> Unless it does something smart like trying to compensate for the latency on the client end

This is actually already done in many AAA titles. The idea is to basically always be showing a prediction of what will have happened in ~70 ms, and update your state model accordingly. Given a datacenter to run the games on and all of the announcements about being able to take a snapshot of the state, this doesn't sound super far-fetched.

But stadia can't use these techniques. The game engine and rendering isn't done on the client side. Normally a network game the client displays the user's actions immediately, and the lag is only seen from other player's perspectives.

On a streamed game, all interactions need a round trip before the user sees any response.

Didn’t Google say something about using AI for this kind of game state prediction at Stadia presentation?

Doing anything where mispredictions result in a few frames of doing the wrong thing escape to the user is going to make games feel really inconsistent. If the game is running at 60fps and you have a system that can predict with 99% accuracy (which seems wildly optimistic) you’re still going to have mispredictions at a pretty high rate.

Not to mention that the game will have to be written especially to allow for the simulation to be rolled back and forth. For example a misprediction triggers a sword swing it turns out you didn’t make that traps the player in a state for a second. You end up giving the player a horrible experience where the game seemingly has a mind of its own. Or you can roll back and interpolate from the bad state back to a good one which would be ideal. Then the middle ground is to rollback and have a cut in the action back from misprediction. Then you realise this is happening every second or two.

Game state prediction doesn't cut it, or rather there's no point to it. The rendering is done server-side - remember they're streaming HD AAA games to televisions and tablets. There's always going to be a round trip between pushing a button and something happening onscreen. Google did say something about device inputs being sent straight from the controller to the server. Maybe this means they will cut down on latency between the input and the server, but it's still a trip back down from the server to the display on top of that.

For a networked multiplayer game, the total amount of network lag is effectively the same, except that the network lag is moved from between client and server to client and "display". Now that the server and client are now just one instance, there is zero lag there.

Except, that's not actually the case as you still have latency between servers.

This is only relevant if you're deliberately splitting the game across multiple servers (MORE work) so that people living in different cities can play against each other and have low latency to their own server. Other than that there's no reason to run one multiplayer game on more than one server.

Not entirely true. Many networked games use client side prediction, and collision detection (at least for collisions against terrain and walls, weapon hits are often computed on the server to curb cheating).

If you have a 100ms latency to a stadia server, then there's a 200ms latency between pushing forward on the stick and your character moving forward. This is not the case on a networked game, the client would start moving your player immediately (even if there is a 100ms delay to send those updates to the server).

Considering that pretty much all multiplayer games in existence perform some sort of local prediction / state interpolation to hide the lag on the local machine, cloud only multiplayer will be considerably worse. After all you can no longer hide the lag locally, since you're not computing anything on the local machine, so the minimum precieved lag will go from 0 (for movement of your own player character) to the RT between the datacenter and your PC :/

I’ve played a lot of Sekiro on XBox One. I’ve played Bloodborne on PS Now and it works great. They have a free trial you should check it out.

Bloodborne on PS Now? That sounds like hell. Plenty of later bosses have very narrow dodge windows. I already have a physical PS3 and PS4 so there's no reason for me to get PS Now. A couple co-workers of mine used it and none of them liked it. It works okay for RPGs, especially turn based RPGs where there is zero emphasis on timing or reactions.

If you're close enough to the server it's possible for network latency to not add much compared to the combined latency from the tv + console + wireless controller. (most modern tvs add at least 20ms, even on "game mode", some add more than 100ms).

It's a different story on a pc with a decent gaming monitor though.

> sounds like

This is why I'm suggesting you try it. Because to be more blunt, you do not know what you are talking about.

Whether or not latency is an issue is hugely dependent on the title.

It's hugely dependent on particular network conditions. I wonder if SpaceX Starlink might have an option to optimize for latency?

It certainly will but it will be premium for high frequency trading.

I'd be quite satisfied with a reasonably priced tier that guaranteed 40ms round trip or better 99.8% of the time. The HFT levels could shave that time off and start stacking 9's for exponentially more money.

Stadia has a "frame token" that can be sent with the use inoit for cases like this.

That's right. If Stadia does what Google says it does then gaming console business faces long term risk. I was quite impressed with Stadia demos.

Reply on news.ycombinator.com