Pixel Streaming and Command Streaming Overview

  • Pixel Streaming sends images using VIDEO technology
  • Read user inputs and run application loop on CPU
  • Output is rendered by a GPU in the cloud – and you need a GPU PER USER
  • When render is complete, it then gets encoded into a video stream
  • Video packets are sent to the user
  • User device decodes video packets to images and you have to wait until the WHOLE IMAGE renders
  • Average costs ~50 cents / hr / user to run a GPU stream
  • Launch numbers for a AAA game today is multiples of 1,000,000s vs CCU capacity of the pixel streaming vendors is multiples of 1,000s at best; there simply aren’t enough GPUs available in the world’s clouds combined to stream to every user (and no one can today present incontrovertible evidence to prove this statement is wrong…)

Command Streaming

  • Command Streaming sends DATA to draw the image
  • Read user inputs and run application loop on CPU
  • Output is drawing commands compressed on the fly – NO GPUs
  • Data packets are sent via any cloud to the user
  • User device GPU executes DRAW commands in real time
  • Choose stream visual fidelity based on each user device GPU
  • Transport uses BBR-style RTT congestion control as well as burst mode chunking
  • Cost To Serve (CTS) shared between Data Center and user device
  • Average costs ~4 cents / hr / user to run a stream
  • Utilizes caching and CDN services – reduced egress
  • Containerized workloads – dynamic scaling
  • Cloud provider agnostic to be physically close to users – less hops
  • Business models to monetize non-player participants
  • Novel applications for interactive 3D social spaces

Scalability and Latency

Polystream can elastically scale to meet any demand, without compromising on delivery quality or speed. Using state replication and lockstep simulation, we can serve audiences of unlimited sizes, without sharding. Whether adding a single user or 1000, the equivalent demand is of just one player.

With command streaming technology, latency is kept low. Cloud commodity machines without GPUs are ubiquitous and can be located close to clients – usually within 20 milliseconds of end users!

Because command streaming doesn’t rely on expensive GPU machines in the cloud, interactive experiences powered by Polystream can run from almost any data center in the world – at significantly lower costs.

Users can access Polystream powered experiences seamlessly, in just a few seconds. Following a URL posted in a chat or social media post, viewers can instantly jump into any 3D experience as a spectator. User URLs can also be controlled, tracked and time-limited to provide adaptable security & management of the spectating experience.

Our technology is designed to complement projects made in existing game engines like Unreal Engine with minimal performance impact to the experience.

You can build Polystream-enabled experiences using the same code base and tools as you normally would. From there, it’s as simple as integrating with our SDK and Plug-Ins as much or as little as you like.