AMD stakes Opteron future on direct connect marchitecture

It's this that allows its executives talk with a swagger about Opty, and brag about its prowess. Such is the wizardry of the chip, ballsy AMD has managed to lever away around 30 per cent of the target market from the bullying monopoly, as the chipsters like to label their competitor. But CEO Hector's not satisfied with that and has set a new target of 50 per cent market share for his boys and girls to aim at.

On an amble around Europe this week has been Giuseppe Amato AMD's European technical sales director, extolling the virtues of the Opteron architecture to all and sundry.

According to him, the direct connect architecture is so plainly more efficient than any Intel offerings that he can see Intel having to copy it eventually. The only thing stopping them is that AMD thought of it first, he reckons. The mighty monopoly would lose face in copying its upstart rival, after having dismissed the technology out of hand, Amato suggests.

The intelligence of AMD's direct connect architecture is apparent in the drive to multi-cored processors. As the name is meant to convey, AMD says it can connect additional cores or memory or third-party modules directly onto the silicon substrate. The chip maker claims its Hypertransport bus has the advantage of allowing direct communication between cores and between cores and memory.

The architecture allows AMD to move from its current dual-core Operton to a quad core one in the same socket. The firm claims the same power consumption of its quad-core offering as the dual-core demands, although it is unable to clearly state how it achieves this feat.

By the time the quad-core Opteron is available, AMD will have also tweaked the cores themselves and it is likely that the four cores will be clocked at a lower frequency than the lonesome pair. This is what AMD previously told us in the very same room in which Amato told us that it wouldn't be the case, before deciding that any clock difference that might prove necessary would be low enough to be deemed negligible. By him, at any rate.

But this is the promise AMD is making its dual-core customers now: buy a dual core from us and when, in the future, you need to upgrade simply prise out the old dual-core chip and slip a spanking new quad core into the very same socket. And, since the power requirements will be the same, you can even use the same cooling system and power supply.

Based on what it knows about Intel's upcoming chippery, AMD reckons it will rule the roost in terms of performance per Watt for the foreseeable future and makes startling claims for the amount it can save its corporate customers on their electricity bills.

A further demonstration of the flexibility of AMD's direct connect architecture is provided in its willingness to allow its friends to bolt their own processing modules onto its silicon.

AMD outlined its Torrenza initiative to us before and this week anounced some mates that have signed up to toy with developing co-processors for the Opteron super socket.

Sun, Cray, Fujitsu Siemens, HP, Dell and IBM said they thought collaboration with the Torrenza Initiative interesting enough to say so publicly. It remains to be seen if anyone comes up with anything but an ATI-sponsored Physics co-processor won't be long in the making, we suggest.

The move makes sense for an AMD up against a giant with far bigger muscles. If anything comes from the initiative it'll be a boost for the upstart in its search for feet big enough to fill larger boots.

It needs some friends now that the gloves are off and it's slagging off the market leading monster at avery opportunity. And in the courts.

And in he shape of Opteron, it feels it has the product to grow into those boots of Mexican leather that are already on order. µ