In my comments earlier this week on AMD's purpose-built "Fusion Render Cloud" supercomputer, I neglected to mention a possible downside for AMD's GPU business. In a nutshell, if this new supercomputer is going to be doing all the heavy lifting rendering-wise in the server, why do you need GPUs in the client?

The issue is probably more obvious when you realize that the supercomputer is being built with essentially the same "Dragon" chipset destined for high-end multimedia PCs. Specifically, it's the ATI Radeon HD 4800 GPU in the chipset that delivers all the nifty HD multimedia capabilities coveted by hard-core gamers and video enthusiasts. And it's not just for supercomputers and desktop machines. On Wednesday, AMD introduced a slightly less powerful offshoot of the HD 4800, the ATI Mobility Radeon HD 4000 series GPUs. These chips are aimed at the notebook market and promise to deliver "a home theatre-quality HD multimedia experience."

But if AMD's petaflop rendering monster (containing 1,000 Radeon HD 4800 GPUs) is truly able to deliver a cutting-edge multimedia experience to low-end PCs, then why buy the expensive box at all? And since the 1,000 GPUs in the supercomputer will probably be utilized more efficiently than in a 1,000 separate PCs, overall AMD will need to manufacture fewer of them to deliver the same computational performance.

Right now AMD is probably more focused on the upside of the Render Cloud. Since any device smaller than a notebook (netbooks, mobile phones, etc.) is not likely to house a CD/DVD drive for multimedia -- not to mention a discrete graphics processor -- the mobile market presents a natural opportunity for HD streaming, without the threat of cannibalizing current GPU revenue.

On the other hand, multimedia notebooks and desktops could get trapped in the crossfire. Users might decide to jettison the pricey GPUs (and DVDs) in favor of streamed multimedia content for the sake of convenience. Not only could gamers stop buying DVDs at $29.99 a pop, they also wouldn't have to upgrade their machine every time a graphics processor came out promising the latest whiz-bang special effects. Instead the cloud would get the upgrade, while the thin clients automagically pick up the new capabilities. So is there room for high-end GPUs on both the client and the server?
 
I guess the answer revolves around the pricing structure of content serving versus media ownership. Presumably content providers are planning to use some sort of subscription service to deliver games and other HD content from the Render Cloud. Since most online games and HD media on the Internet are currently available for free, there is a lot of pressure to keep prices low. But how low?

A model that already exists for something like this is Amazon's Video On-Demand, an online service in which digital video content can be purchased for lifetime ownership or merely rented for 24 hours. The content can be viewed offline (downloaded) or online (streamed) and can be run on a PC in a standard Web browser, or on other devices, even TVs. Using the Video On-Demand service, renting the "The Dark Knight" movie for a day costs $3.99, while buying it costs $14.99. Alternatively, if you want to own the movie as a DVD, the price is $20.99. The Video On-Demand means the DVD player -- the content renderer, in this case -- has become superfluous.

Obviously, content providers and chipmakers don't have the same interests. Game companies and other media developers will find a way to make money from their intellectual property, even when its form or distribution changes. But if AMD is going to serve up graphics computation on demand, it's risking making the client-side GPU hardware redundant. Of course, if the company can convince users that there is unique value for both client- and server-side GPUs, then problem solved. I guess that's why God made marketing departments.