AT CES this year both "leading" graphics vendors have gone on the cheap to bundle two of the current offerings on one card and claim a new generation in performance.

So, each GPU on such multi GPU card, whether it's HD3870X2 or GeForce 9800GX2, has its own memory which simply means lots of data duplication and wastage. The two GPUs are connected over PCI bridge chips, which means extra latency; sharing the outside PCI-E x16 v2 link - read: possible (but not seen too often) contention between the two.

Did we mention that current software is hardly being optimised for all that SLI and Crossfire stuff? In most cases - GPGPU excluded - one fast GPU is still way better than two slower GPUs at the same price and power consumption.

Then, we come to the physical implementation. DAAMIT at least had some common sense to put both GPUs and stuff on a single card. Therefore, if someone wants to do so, we could still have a cool yet efficient, one-slot waterblock like those from Vadim Computers in Old Blighty, for instance.

That means, a dual-card, quad-GPU CrossFireX on an Intel X48 or Seaburg-chipset mobo - or the odd Phenom around, for those who just don't give up - won't block all the remaining I/O slots.

Well, at least the product is correctly named: 3870X2 really is TWO 3870s put together - even the clock speeds are expected to be basically the same.

On the other hand, Nvidia created a double-decker PCB monster not unlike the old, infamous 7950GX2. The (mis)named 9800GX2 is in reality just two "new" 8800GTS-512 units put together, giving each "half" its own PCB.

Right, there go any hopes of that thing having single-slot cooling options, and therefore any hopes of us endorsing that to anyone, anywhere. Where are all those D9E, D9P and other new-generation chips? And, with such odd naming scheme, how to compare the 9800GX2 with a future 9800GTX then?

Now, why should Nvidia bother at all? DAAMIT was slow, they still can't beat the over-a-year-old 8800GTX or Ultra yet. Why not just take a nap, rest on their laurels, pick up extra profits and, occasionally, spin a process shrink of a few GPUs with some (incomplete) fixes like HD playback, for instance.

Sure, but DAAMIT right now lives off those GPUs and their management, no matter how inept, will not - cannot - let the GPUs fall behind any further. Plus, looking at the K10 woes, methinks the GPU fix will be far easier.

Nvidia should be concerned - not just because the dual 3870X2 will make for a potentially more elegant watercooled quad-GPU solution, but because of what the R700 will bring in, once its multiple MCM-ed GPUs don't need to use PCI-Express to talk to each other - a critical GPGPU performance jump enabler. Nvidia, needs to speed up the launch of high-end D9 series, rather than sit on it.

Nvidians can continue to snooze, but things are coming to their right place - look at DAAMIT's FireGL profi card improvements and how they shook Quadro, as a prelude of the things to come.

And un the centre, Intel is not slowing down, neither on Nehalem, nor on the Larabee. The unhealthy airs in the current GPU market, from the product refresh delays to the GPU vendors' attitudes, will only help push the Taiwanese OEMs even further towards Intel's GPUs, across the board.

And, sad to say for the competitive environment, Intel has learned - the hard way - to listen and adapt quickly. µ