AMD talks about the incoming platform

Aug 23, 2007 10:09 GMT  ·  By

AMD looks forward to the integration and merging of the two most complex computer parts, the central processing unit and the graphics processor, as this step would bring better graphics quality and capabilities, while presenting end users with the ability to run more parallel centered software applications that are better processed at the GPU level.

According to Phil Hester, AMD's chief technology officer, that spoke about the company's plans to unite the two previously separated computing platforms, the Fusion chip should hit the market sometime during 2009 at which time the entire market will be looking for a graphics chip that can deliver great performance but at an affordable price. Phil Hester compared the AMD's decision to built the Fusion chip with the very inspired and profoundly revolutionizing decision that Intel took when it integrated the floating-point processor into the 486 chip. While integrating a new unit in an already well-established design can lead almost surely to higher manufacturing costs, if enough applications and customers are expecting and welcoming the performance boost, the increased costs are almost immediately covered.

The plan behind the Fusion architecture is not to replace existing and future high end discrete graphics cards as it will not be able to deliver a very high 3D performance. The performance gain will not come as the CPU helps the GPU but rather the other way around as graphics processing units are capable of an increased level of parallel computing that normal and general use processors can not yet match. According to the news site News.com, there will be a series of advantages in terms of general performance gains from the Fusion architecture, as long as developers can and will write code that implements the Fusion extensions.

Now, Fusion is little more than a name on a roadmap and there are still a lot of design and concept decisions to be made. For example, the exact type of memory the new chip should use: typical DDR or video memory, as both solutions come with distinct advantages and disadvantages. Another problem is that the new hardware architecture should be quite open source oriented in order to attract more and more software developers that would write applications with support for its parallel-friendly setup. Hester said that integrating all kinds of features on the processor die does not make too much sense but some of these technologies could have a great performance impact while reducing the overall costs for end users.