Cloud Computing

I just read this interesting article about AMD's plans to build a 1000 GPU supercomputer to research cloud computing for the use of gamers and 3D animators.

http://www.custompc.co.uk/news/605374/amd-plans-supercomputer-with-1000-gpus.html

I'm considering the long-term implications for a product like Second Life, where centralized rendering could allow tighter synchronization of animations and more common representation of in-world object relationships. Potentially, this could lead to real-time ray-tracing as well. With enough nodes and bandwidth, of course.

In a few years, I'd expect this kind of architecture to be a lot more common, with more powerful nodes and a lot more of them.

Pulling all the major data back into a central server will also eliminate or severely reduce issues about IP infringement, as no one would ever download the raw textures and shapes currently needed by the current clients.

If you think network lag is bad now, imagine what it would be like if you were trying to stream a 1280x1024 video of SL to your screen.

And we could have realtime raytracing now if nVidia's chief scientist didn't have a burr up his butt about it.

Is it really practical to offload rendering onto a remote party? I'm trying to get my head around how that would work but it really seems like there would be a big disconnect between me moving my mouse around and the ideo being delivered over the wire. It would all seem very... squidgy because of the half second delay between your mouse move and the video arriving.

Elanthius, I'm proposing that Linden Lab themselves might someday bring all rendering back to their servers. As I said, this has benefits for synchronization issues, a common and consistent rendering "style" regardless of client capabilities, protection of intellectual property, and a drastically simplified client that is easy to maintain across multiple operating systems.

But, as you point out, there may be simple laws of physics that get in the way; while the computing power might be available within a few years to simultaneously render 50-100 thousand views in a virtual world at once, a 100 to 500 millisecond delay that's typical with Internet communications might be enough to spoil the experience. I wonder what AMD's approach to this problem will be.