About us

Scan Windows 7

Graphics gets cosy
As soon as their merger with ATi was completed, AMD wasted no time in announcing their first graphics product, Fusion.

Fusion aims to move integrated graphics away from the chipset and onto the CPU itself. Much like a dual-core processor has two separate cores, a Fusion processor would have a separate core to handle graphics.

Click to enlarge

AMD has extensive experience in moving core technologies away from the chipset, as both Opteron and Athlon 64 processors have the memory controller built onto the chip. This move has given them an advantage in memory bandwidth, which they still manage to hold over rivals.

However it's a bit early to get excited and not just because AMD say the first Fusion parts won't appear before late 2008/early 2009. Processors with integrated graphics will be aimed at the value market; one which is traditionally an Intel stronghold.

Probably most interesting of all is the promise that Fusion won't just be about integrating graphics cores onto processors. AMD mentioned other specialist co-processors which would handle Java, XML and media applications.

While AMD were tight-lipped on how specialist co-processors would be chosen for inclusion into future processors, Phil Hester, AMD Senior VP, said that co-processors would only be included when "a large percentage of applications will make use of it." This makes sense adding a co-processor not only increases cost for the consumer, but for AMD too.

Click to enlarge

AMD also revealed a modular approach to processor design that they say allows easier customization of processors to different market segments. Using integrated graphics as an example, Phil Drebin, CTO ATI, showed how a value orientated processor may only have one graphics core integrated on chip, whereas a processor aimed at gamers may have multiple graphics cores. The aim for AMD is to have a toolbox to delve into, mix and match blocks in order to create a processor for their target market.

With so much talk of integrating graphics right onto the processor, AMD were keen to point out that discrete graphics - that's standalone graphics cards - weren't on their way out. Hester stated that teams working on discrete graphics "haven't changed in size". Furthermore, he sees high-end users' needs being met with standalone graphics cards, as integrating complex graphics solutions would increase processor pricing.

Fusion seems a fairly logical step from the Torrenza initiative announced a few months back. The ability to build specific co-processors onto the CPU is nothing new; maths/floating point co-processors have been implemented on the processor for a couple of decades. However, the ability to tailor a processor which is optimized to a specific task is quite enticing.

While AMD are more than happy to do the legwork with on-chip graphics, thanks to their merger with ATI, other co-processor technologies may have to be developed by third parties. AMD's job would be to implement silicon for that technology when large customers such as Dell, HP and Sun demand it. This isn't a bad thing and is common when companies look to the open source community for technology.

With Windows Vista's user interface having many 3D enhancements, it would be very nice to see those being handled by some lower specification graphics processor, such as one implemented on chip. This could leave a dedicated graphics card free to handle the more demanding applications such as games. Drebin cites Vista as "setting the scene" for applications such as word processors to become 3D.

Clearly that's a high-end scenario, however integrated graphics on CPU will have other benefits: lower cost and power efficiency will be greatly enhanced thanks mainly to avoiding data buses. From a manufacturing standpoint, motherboards will be cheaper to make and offer the consumer a cheaper upgrade path should PCI-Express be present on the motherboard.

However it's not all sweetness and light. The fact is Fusion represents a major step, especially in implementation. AMD see this as the single most important evolution, since the AMD64 architecture which they first announced in 1999. That took almost four years to materialize and this time AMD want to have processors out in half the time. The biggest problem will be achieving high yields in fabrication, enabling them to price Fusion processors competitively.

The incorporation of on-chip graphics will be the first real indication of how the AMD-ATI marriage is going to work. Taking on Intel on both processors and on board graphics is a tough task, we just wonder whether Intel will announce such a bold move in order to retain its market share.