| walsufnir said: Removed for our sanity
"Cloud Compute execute code or jobs like your CPU/GPU would natively on your console. Think of how the Cell processor works. Developers create multiple jobs which are just snippets of code that they send to the different cell processors to work in parallel. The Cell processor works on each code independent of the other processors and sends back it's results. From here the main processor takes the finished code and deliver the results depending on what was calculated like, AI, physics, lighting you name it. With cloud compute, MS is able to leverage thousands of servers each having a number or processors and each processor able to execute multiple threads. The way that MS cloud platform works is that all of these resources are virtual. They are not tied to any particular hardware for fault tolerance. " First, no, the cell doesn't compute independently, because they have common ressources the spus have to share. But anyway this is not what you can "off-load" because bandwidth and latency are magnitudes better than whatever internet you will have. What you are trying to say is not possible because the computing units need data fast - in time and amount. If it would be that easy they could also build a system with a pentium2 and offload everything to the cloud. Local hardware wouldn't be necessary at all (even with dedicated hardware). Also 300k servers would be way too less for the amount of users they are expecting.
People have stated that why would MS go down this route when you have games able to do the calculations upfront. As I was doing a little digging on my break. I stumble on this intel project to do ray tracing using cloud compute. The article presents some interesting ideals on the type of rendering techniques that could be performed on the cloud that you would not see within your current or even next gen console. Here is the link http://software.intel.com/en-us/articles/experimental-cloud-based-ray-tracing-using-intel-mic-architecture-for-highly-parallel"
As I understand it they are talking about the benefits of calculating of calculating in the cloud and offering a stream of this to the user: Gaikai/Onlive. This is different to what MS wants us to believe.
|
OK, maybe I should not be so general. The SPEs can run independent of one another as the PPE is the part that schedule task amoug them. The SPEs do not share resources as they use DMA to directly read and write their data and rely on their local memory. The key with cloud compute is that you only execute an API and the processing is done on the servers. There is not need to feed data from the client PC besides a few parameters and the cloud does the rest.
If it would be that easy they could also build a system with a pentium2 and offload everything to the cloud. Local hardware wouldn't be necessary at all (even with dedicated hardware). Also 300k servers would be way too less for the amount of users they are expecting.
Actually, what you present is exactly where they are going with the technology. As to the amount of servers, this is the number MS stated would be available at release. This will not be the number in 2 years from now.
As to the Intel article, I already stated in another post how this is totally different from Giakai only being he same in how a scene is delivered to the client. The key is that with Cloud compute, a developer could potentially use many different rendering techniques supported by the platform while Giakai can only stream the game its current hosting.
