Intel and AMD share halls but not goals

IT WAS INTERESTING to note that, this Computex, Intel and AMD not only shared advertising space on the same facade (see related story), but also had their showcase booths in the same Hall 4 (TICC convention centre).

However, the attitudes couldn't differ more - while AMD was showing its existing product line, with minor updates like long expected announcement of 64-bit dual-core Turion, Intel was focusing on its upcoming Core 2 Duo, mostly Conroe and Woodcrest parts first - oh yes, there was one Montecito demo system there as well.

The most interesting stuff this round was not so much the array of goodies shown (after all, there are hardly any secrets about AM2, Conroe or Woodcrest left to discover), but the one-to-one corner chit-chat discussions with those in the know, whether from Intel and AMD, or from the local OEMs. And, from these talks, I sense smell of war coming up - and one shouldn't doubt such a sense from anyone born on the tough terrains of what was once called Yugoslavia.

The trenches for the summer battle are already dug deep, way before this show: Intel's Core 2 Duo attack force was everywhere, with Conroe-ready boards on every vendor's booth - something that wasn't really hard to prepare, since most vendors' boards using Intel 975X or Nvidia Nforce chipsets will run Conroe with at most a BIOS update. The dark horse that should really propel the FSB-based Conroe XE performance, whether in standard or overclocked operation, is the up and coming ATI RD600 chipset - assuming, of course, that the flushed-down rumour of AMD-ATI merger doesn't suddenly resurface as a fait accompli in the meantime.

Running at the ATI demo lounge in Grand Hyatt, RD600, according to its creator, should outpace both Intel and Nvidia chipset offerings for Conroe in few key areas, from far better FSB electricals and overclocking (2GHz reliable FSB operation is a target, a nice synchronous match for dual-channel DDR2-1000 memory), to lower-latency memory controller, and, supposedly (but no confirmation), over 40 PCI-E lanes to cater for dual PCI-E X16 CrossFire and some more I/O on a side for, say, physics GPU or PCI-E X4 SATA RAID controller, for instance. So, it is interesting that what was rumoured to be AMD's takeover target actually creates the best chipset platform choice for Intel's new desktop CPUs - and maybe, if ATI goes ahead with a dual-FSB workstation system chipset for Woodcrest, the best dual-socket choice too.

With this kind of chipset, and with tacitly acknowledged high overclocking margins if using good cooling (with Corsair Nautilus 500, AMD FX62 2.8GHz only goes up to 3.1GHz, but Conroe 2.66GHz goes to 3.6GHz stable without even a voltage change!), the benchmarks on the new platform might look darn dangerous for the AM2. Several Taiwanese big names also confirmed initial good speed test results with the Conroe, but no one was supposedly allowed to openly show the overclocked units on the booths.

AMD, in defence mode now against advancing enemy forces, ensured that those same vendors peddle the still fresh AM2 platforms everywhere - since they relegated the chipset platforms to the partners, it was a mix of Nvidia and ATI-based entries around the show floor.

With Turion64 X2, AMD also took the advance shots at Merom, which is still some two months away - they perceive the "freedom of choice" for the chipset and wireless circuitry as a major plus compared to the current and upcoming Intel offerings. However, AM2 is still left wanting, as, besides the DDR2 enablement and somewhat lower power, there isn't anything new for this year. The rumours floating around the show floor see it only as a passing downturn, as AMD has supposedly settled its 65nm process, and limited units should appear before yearend - at least to claim the 3.2GHz 'FX-66' dual-core speed peak, if nothing else. If that won't work, well further tuneups of 90nm process should anyway bring them towards that goal.

Then, the '4x4' dual-socket initiative seems to rely on HT to deliver some of the benefits of nearly-ready Kentsfield to the 'enthusiast' PC community, without creating new parts. I haven't been the VIP enough to see it this time, but one thing I feel it will surely do it - push forward the Kentsfield lauch. Am I wrong here, Intel?

Also, since K9 and K10 were gone for a variety of reasons, K8L might end up to be more than just quad-core K8 with faster FPU and L3 cache (and yes, larger address space for the supercomputer use). Our guys have mentioned the quad-issue upgrade possibility, and one of the AMD friends on the floor just smiled at the mention of it. K8 architecture has enough execution units to handle a quad-instruction input, even if it means four complex instructions at a time. So maybe, after all, it may come closer in per-clock efficiency to the Core 2 Duo / Quad.

But again, this is all still vapourware that no one in the open can see or touch yet, and we could talk about Nehalem core too then - after all, it is due just a year after K8L. An interesting concurment by more than one OEM party here in Taipei was that Intel, if they want so, could easily support HyperTransport 3 in their 2007 CPUs, and still use CSI as well around end-2008, even overlapping the two interfaces, just like PCI-X and PCI-E for instance. The gain of having HT3 early wouldn't just solve the problem of quad&eight socket market competiveness, but also let Intel make use of all those HT-based north bridges by Nvidia, ATI and Serverworks for some nifty workstations and servers.

Oh yes, by the way, I almost forgot the VTF - VIA Technology Forum shared the same third floor of the aging Grand Hyatt together with MSI and Gigabyte booths. The order of visitor density, per my perception, was MSI 3 : Gigabyte 2 : VTF 1 - and, if you go around the booths, you'll have to search for VIA-based solutions like for a needle in a haystack . VIA was my favourite independent lone ranger in the PC CPU & chipset market, but, the emptiness I see this year truly worries me - I hope that, by next Computex, VTF doesn't become a WTF.