Partners and Our Thoughts
The truth of the matter is that most of this technology is available in other areas and we are eager to see how NVIDIA plans to differentiate their GRID service. One way for them to do so is with Kepler GPUs and the power efficiency benefits it can offer to server farms that are doing the streaming for Gaikai, etc.
Increased density of compute power is huge for these companies and if NVIDIA can in fact raise the number of game streams possible per server node by a factor of four it would be a tremendous benefit. By enabling 4 GPUs for game streaming purposes in a server NVIDIA claims to improve the power consumption for each game stream as well by 50% – going from 150 watts to 75 watts.
NVIDIA has built a compute card exactly for this purpose known as the NVIDIA GeForce GRID Processor, a dual-GPU card that includes a total of 3,072 CUDA cores and 8GB of memory. Likely this is a pair of GK104 GPUs (similar to those seen in the GTX 680) but with a TDP of 250 watts it should be running at slightly lower clock speeds than the recently released dual-GPU GTX 690 card. This is also likely the same design as the newly announced Tesla K10.
The goal of not just NVIDIA’s new streaming technology push, but all streaming gaming companies, is to make the game console a virtual device. Playing games on your TV (without the need for a physical console), on your tablet and even on your phone with graphics that you would only expect to find on a PC is a lofty goal that would benefit gamers all across the world by bringing down costs to entry.
NVIDIA isn’t going at this alone though and as you would expect of a company with this much gaming cache, they have lined up the best in the business. Online game streaming services have signed on including Gaikai and OTOY, virtualization technologies from Microsoft, Xen, Citrix and VMWare are support and system OEMs like Cisco, Dell, IBM and HP will be able to produce systems with the GeForce GRID processor.
We are really just starting to learn about NVIDIA’s new GRID technology and because of that I will leave our final verdict for when we get some hands on time with an implementation of it in a real-world environment. I have still many questions about what NVIDIA is actually doing to improve the latency problem in online streaming gaming and we only have part of the answer today. While the H.264 encoder in the Kepler GPU will definitely help on both ends of the network connection (server and client) the truth is that Sandy Bridge and Ivy Bridge CPUs have had that capability for some time – what does NVIDIA offer above and beyond that?
Improving the latency of game time is also interesting – yes Kepler can render a 720p or 1080p image much faster than previous GPUs used on streaming servers today, but how can GPUs are increase programmatic "game time" internal to the game engine needs to be explored.
NVIDIA is definitely treading into new water here by not only supporting the streaming gaming solutions but also pushing them. It is possible that this would backfire on NVIDIA and users would decide to subscribe to Gaikai rather than buy the new GeForce GTX 680. Another theory is that NVIDIA has missed out on all of the next-generation console licenses and will instead take the battle to the entirety of the console market by offering "console quality" gaming through GRID.
We have much to learn about the improvements brought forth by GRID and we’ll be sure to share them with you soon.