Machiavellian said:
Yes, I read the articale all the way through. Do not forget that they are not using a compression scheme like H.264 but Dxt. They are not using the more bandwidth/latency technique like the Tiled Based rendering but the scene based one. They do mention the different compression schemes and do remember that the X1 has dedicated hardware to perform those actions. Also the article is a proof of concept and more than a year old. I use it as a means to show the different rendering techniques that could be implemented by using the Cloud that is not realistic or possible on current or next gen hardware. Also this is not similar to Gaikai as Gaikai host a game instance on a server. Gaikai cannot split different processing of a game code like this solution as Gaikai can only just execute a game just like you would on your PC. The games are run as if you ran them on your PC and they are not distributed amoung other servers for load balancing or distributed work. Also Gaikai cannot change the rendering, physics, lighting or AI of a game which this solution would. The only simlar part is how the fame is sent to the client machine as a compressed video. The Intel implementation is actually running game code which is the rendering peiece on multiple servers. Its distributing the work amoung multiple servers. As the artile states, it takes either a piece of a frame and distribute them or it can have each server process individual frames. This is eactly the type of setup MS is talking about. MS cloud compute can take individual code and execute them amoug multiple servers. It can split the task and create new instances to process the work. MS cloud compute can be the rendering engine just like this Intel demo or it can process, other parts of code. All of this is done based on the AI Intel created for their rendering engine which probably would like the Orleans setup within within Azure. Think of this demo as an example of MS cloud compute actually rendering a scene within a game using cloud compute and ray tracing. It doesn't stop there as mentioned witin the article, other highly CPU intensive task like Voxels or point rendering could also be done using cloud compute. I guess the key is that Developers will have a host of options on the rendering from besides the ones mentioned within the EG article and given into the hands of talent developers could prove impressive. There are other articles on doing raytracing over the cloud including MS Azure being the platform for Pixar rendering farm. |
Yes, I get where the cloud computing, or rather distributed rendering comes in. (essentially an sli setup but on seperate servers)
Once you're rendering on the server you can distribute it as much as you like with all kinds of exotic and new rendering techniques. (Although having 4 servers rendering at full capacity per 1 client is no way to stay in business)
However it has the same drawback as Gaika, added lag from input to end result. There is no direct feedback possible as the client has to send the input to the server to receive the corresponding result much later.
I'm trying to think of ways how cloud compute can increase the performance of games without introducing lag. Basically anything the user can manipulate with an immediate effect (view point, interactions with the world and npcs) has to be handled locally.
Their idea of using it for a surveillance station is something that works, not time critical. The server can make beautifully ray-traced custom made cut scenes, all party members present in their correct outfits. (Assuming you have enough bandwidth not to get ugly compression artifacts)
Other then that I don't see how you can get away with cloud based rendering without sacrificing response time.
