I'm working on building a game streaming cloud service at the moment.
In the long term, I believe running your own servers is going to be the future, because otherwise there'll always be license issues, e.g. when you want to play your gog.com games on GeForce Now. Or indie games or business apps in general.
Also, all those services prevent you from sharing with friends for their business reasons, meaning absolutely no coop or splitscreen.
Anyway, the key for making such services work is custom ultra low latency udp protocols. I'm going with nvenc hardware encoding, cuda for data wrangling, and a boost::asio based c++ core for the network layer. That, and controlling packet loss, for example by self-throttling and spacing to avoid overflows at intermediate relays.
BTW, I'm surprised by the bandwidth numbers in the article. Even fast explosion heavy games like BroForce work reasonably well with 5mbit/s if you use h265.
Microsoft and Sony have a large catalog to offer, Amazon and Google have large data centers, a lot of money to throw at publishers, and a good path to integration with streaming (Youtube and Twitch).
Geforce Now has shown the hostility independent efforts will have to face.
I also believe the target audience is mostly users who can't or don't want to afford dedicated hardware and will want a good collection of games with a subscription.
I don't plan to compete with those huge services and their catalogues.
Instead, I'm building a platform as a service. My ideal customers will be the software publishers that want to run their own streaming cloud. There's a lot of demand for enterprise solutions that allow you to demo things online. With a regular unlimited demo version, you have the risk of people trying to crack it. If the demo is a streaming service, then you can use everything, but copy nothing.
In essence, a high performance streaming solution is the perfect copy protection for pricey enterprise apps.
sure for gaming, but why would i run my enterprise software on your pc wherw you can see my propertary data. does not make sense except gaming and some other smaller things.
I think it makes a ton of sense for things like EDA software, they have to run on beefy machines that are shared between users and the user experience over thinlinc / X11 forwarding is typically pretty atrocious. If you sell that solution to Cadence they would be super happy probably.
There is a lot of appeal in "play anywhere" alone. Such software gives you the ability to better utilize your hardware by playing from your laptop, friend's house or phone. Also, imagine only needing a weak device as thin client and doing 3D or photo editing through your main computer.
This sounds great, I would love to talk with you more about this if possible. The technology used at Stadia is streaming using open-source Vulkan graphics and it is Linux based, this gives you a lot of leverage since its open-sourced.
Just wanted to wish you luck here. There is a huge need for competition in this space. Moonlight and Parsec are both good, but seem to be stagnant performance-wise.
I find moonlight unbearable to setup and parsec tends to be limited by my low upload speed so that it looks like a blurry mess for my friends. In one top down racing game, it got so bad that my brother literally couldn't find his car onscreen.
After trying out Stadia and GeForce Now, what I personally want is a high performance data center like what they have, but with much more control for the user. There's no reason why I should be limited to only play games in their catalogue. Because that denies me pretty much all indie games.
I seriously hope you consult an IP lawyer on this. There was a lawyer on youtube a few months ago that pointed out that excuse doesn't necessarily work as a defense when it comes to copyright infringement. https://youtu.be/JKjxfDJXV1E?t=1084
Except that h265 is terrible for encoding, you'll burn so much CPU juste to encode the video. ffmpeg with a zen3 5900x encode couple of frames per seconds ( but it doesn't have any hw encoding ).
nVidia’s range of cards ship with dedicated hardware encoders; anything from the last half-decade can encode h264, and the 2000 and 3000 series cards can do h265 fast enough that a 4K stream renders frames and transmits across my local network (over ethernet) in less than 16ms. I’m not sure if AMD is shipping similar hardware.
One stream per card, as far as I'm aware; however, given that we're talking about streaming more demanding games, you'd need dedicated hardware for running the game as well.
I think nVidia built NVENC with an eye towards their own streaming service (GeForceNow), with an earlier version (GameStream, which lets you stream locally to a Shield device) as a test-bed.
In the long term, I believe running your own servers is going to be the future, because otherwise there'll always be license issues, e.g. when you want to play your gog.com games on GeForce Now. Or indie games or business apps in general.
Also, all those services prevent you from sharing with friends for their business reasons, meaning absolutely no coop or splitscreen.
Anyway, the key for making such services work is custom ultra low latency udp protocols. I'm going with nvenc hardware encoding, cuda for data wrangling, and a boost::asio based c++ core for the network layer. That, and controlling packet loss, for example by self-throttling and spacing to avoid overflows at intermediate relays.
BTW, I'm surprised by the bandwidth numbers in the article. Even fast explosion heavy games like BroForce work reasonably well with 5mbit/s if you use h265.