AMD Torrenza will be key to keeping Intel at bay

I think that too many commentators have been too fixated on the present when the chip maker's future has so much promise. Yes, AMD's current plight is not a happy one. But the company has experienced harder times with a far worse technological base than is the case today.

I believe the technological foundations that the company requires to be successful are in place, which should help the chip maker to bounce back. So from my perspective it's far from the end of the world. This piece will hopefully make that abundantly clear.

The current state of affairs
With its Core microarchitecture technology, Intel has launched some of the best general purpose x86 processors to date. But if any of the chip giant's future platform technologies - Geneseo, common system interface (CSI) or on-chip memory controller should fall short of expectations, or worse, not make it to market at all, that could leave Core short of critical armour to fend off an AMD64 attack.

AMD has already claimed that its 65nm quad core Barcelona chip will be up to 50 per cent faster than the chip giant's Clovertown counterpart. But Intel may counter with 45nm chips in Q1 next year, so that claim - if true - might be short-lived.

Because of what the chip giant delivered with Core, AMD doesn't want another wake-up call like that. So what could the company do at the platform level that would maintain performance leadership not just for months but for years to come?

Two sides of the same coin
The word that summarised the midyear AMD technology analyst day held last year was consolidation - build upon an already solid foundation. The chip maker conveyed in its presentations that its technology leadership would be maintained. But in the light of Core the company was clearly shooting from the hip.

The major analyst day announcement for me, that could have far reaching consequences for Intel, was Torrenza - accelerator enabling technology that connects into the AMD64 fabric. Torrenza could facilitate the acceleration of FLOPs, XML, gaming physics or whatever the market demands. AMD claimed that Torrenza enables its partners to build the most exciting solutions in history. With its ATI acquisition, the company is now a co-processor accelerator player as well.

At the autumn Intel Development Forum (IDF), the company announced its response to Torrenza: Geneseo, which is a PCI Express (PCI-E) technology enhancement that also maintains backwards compatibility. But that was only a proposal - the technology still has to be delivered. The chip giant also said that it would license its front side bus technology to co-processor companies, which would enhance performance by coupling the accelerator closer to the CPU.

Absent from the Intel Geneseo announcement were any performance projections, which is strange since PCI-E is already widespread and the second generation of that interconnect should have year-end product and application availability. The Intel Geneseo web page did link to the IDF Geneseo keynote. But that was lacking on the performance front as well. That page also linked two Geneseo PDFs. But one document opened with an error message and the other was missing performance estimates as well. In the absence of performance claims, the chip giant still has dozens of companies on board to ensure that the proposal does become reality.

Seeing how both companies announced their accelerator enabling technologies the differences couldn't be more stark. AMD is touting Torrenza as the best thing since sliced bread. Intel on the other hand is understating its position. Why would the chip giant do that? Somebody at Intel must have extrapolated what Geneseo is capable of delivering. The chip giant's stance suggests that Geneseo will fall short of Torrenza's projected capability.

It could be that Intel, IBM and the other Geneseo consortium members have agreed for now to not make performance projections public. But since AMD and its Torrenza partners have already done that, and more, one would think that the chip giant wouldn't be happy to leave Torrenza basking in the limelight.

Because of Torrenza's potential, I believe that this enabling technology will prove as important as the AMD Opteron launch itself. I also think that as AMD64 proved Netburst's demise, Torrenza will ultimately limit the success of Core.

Now some may think that, like AMD, I'm just shooting off from the hip. But before anyone comes to that conclusion let's review two predictions that some may have judged as off the scale.

I said in a June 2003 piece: "If Intel's 90 nm Netburst products don't pack anything more substantive than more cache and higher frequency, then from where I'm sitting, Opteron's server performance lead could be maintained for at least the next year and through the 90 nm generation."

That prediction was fulfilled, which gave AMD an overall multi-year advantage. But it's the chip giant's 65nm offerings that are now giving the company such a hard time. I also made a later prediction that Dell would adopt the AMD Opteron processor - you know who wore the bunny suit at last fall's IDF.

Torrenza's platform diversity
AMD describes the HyperTransport HTX expansion slot as enabling first generation system-level co-processing that is available today. Co-processor accelerators could be designed for PCI-E or Opteron socket use. These devices could also be designed as part of a chipset or even integrated into the CPU.

A dedicated co-processor as part of the CPU itself makes for interesting possibilities - imagine a quad core design that has the real estate space of one core dedicated to just co-processing.

The chipset could prove cheap enough to make it a viable, volume desktop solution. As an example, an Nvidia single chip design suggests bulldozer-like qualities. Motherboard manufacturers could design in a cheap board interface - the HTX slot could be used - so that enhanced versions are optional.

Tomorrow in part two, performance claims that AMD and its Torrenza partners have made will be discussed. µ

See Also
Looks like Charlie will have to wear bunny suit
Intel's Geneseo makes no sense at all

L'INQ
AMD Torrenza
Intel Geneseo