When back in December 2009, Intel announced it would be shelving the Larrabee microarchitecture as a consumer product, it didn't actually say it offed Larrabee completely. No sir.
So how do you wipe the GPGPU egg off your face and turn it into a potential win for the company? Well, you keep working on it without people looking at your work. As some may have noticed over the past months, the company has slowly been building up steam for the Next Big Thing after Larrabee... which is none other than Larrabee.
Well, it seems Intel might have a cunning plan after all, this Next Big Thing, it turns out, is sticking the chip on the other end of a very narrow pipe (by graphics standards) called "The Internet". Instead of giving Nvidia and AMD a run for their money by building new and shiny graphics architectures, they simply make the (3D) graphics component entirely unnecessary.
Intel has morphed Larrabee into what it is calling now Many Integrated Core, or Intel MIC for short. This is a HPC initiative to put a huge amount of GPGPU power on a single card, ie: Knights Ferry. Knights Ferry will be followed up in a couple of years time by Knights Corner, which will likely be a computing cluster, based on Intel's general description of the tech.
Spot the difference: Larrabee vs. Knights Ferry
Placing the GPGPU-based rendering on the other end of that narrow pipe will allow users to run virtually any type of compute-intensive application from any computer anywhere in the world. The applications for this can go from no-nonsense high-performance computing to things like games - in a way similar to that which OnLive came up with a while back. The difference being that Intel thinks a single Knights Ferry card will be able to get a lot more work done than your average GPU.
In essence this is server-side 3D, leveraging whatever good Larrabee brought to high performance computing, and applying it to just about any application. Of course there's an economic model to be built on this... Intel doesn't do these things just for kicks. Intel has to recover a lot of money it's invested, so the first application goes to supercomputing /HPC, a terrain that it is very familiar with, and where Nvidia and AMD have made some progress with GPGPU computing, as of late.
Intel is now turning towards Cloud-based applications, if we are to believe the features of their Single Chip Cloud computer (SCC). Essentially it is a cloud... condensed into a single box. Right now the SCC meshes together 48 cores, or 2x24 if the data we've seen is correct and churn out a level of computing power the world is yet to behold. Something that would fit in perfectly with security apps such as those bearing a certain McAfee logo.
However, as IDF starts tomorrow, Intel thought it wise to cobble together some demos to demonstrate the awesomeness of piped 3D graphics, namely, Wolfenstein on Ray Tracing (who would've guessed). Intel has been working with Daniel Pohl and you can find samples of the RT version of Wolfenstein at WolfRT.de.
We know this sounds wrong (but it looks good). Gaming enthusiasts and programmers have been wrestling with bandwidth for years - and the demands have become progressively steeper year by year. Now all of a sudden you get to pipe data back and forward and play a FPS fave over the web. Did we mention... in realtime? This is what Intel is promising, realtime FPS gaming over a narrow pipe.
It looks good, but is it doable? We'll find out for you during IDF San Francisco 2010. We're as interested as the next man in seeing the practical effects of Larrabee / Knights Ferry, but we'll definitely be looking to see the reaction from AMD and Nvidia, as these steps are all over their toes.