Agreed, the GPGPU movement is more than graphical outputs for home users, it's about the computational power these cards afford for accelerating general mathematics (hence the first "G"). Computations that used to be possible only on supercomputers or highly-specialized hardware is now "easy" on GPUs. The catch is that us programmers need to be able to access that compute functionality -- in which case, I'm a big fan of NVidia for their C-based API, GPU libraries, support of research into GPUs as computational platforms, and educational efforts.
The bottom line: even if NVidia *might* be (slightly) slower at certain operations, the computations are going to get done much sooner than on my CPU. ATI isn't going to get anything done if I can't program it easily, so any (perceived) speed benefits are moot.

For reference, that "special" camera is something called a "plenoptic camera", which has been around for a couple years (see, eg, Mark Levoy or Ren Ng or Ramesh Raskar and their "lightfield cameras"). It's recording a couple different viewpoints simultaneously, one from each fly-eye lens (as the translation calls 'em). That means that the reconstructed optical field has only a few discrete viewpoints -- enough to give a 3D effect with the holographic reconstruction, but it's not yet the complete optical wavefront from the object.
The poor clarity (especially noticeable with the text) is partly due to the tradeoff between number of viewpoints versus number of pixels per viewpoint and probably due to the fact that "3D" is in the defocused background... proving that it's 3D.
I'm also going to posit that the viewing angle is related to the size of the (SLM) LCD pixels, probably around 10 microns to give a 2 degree viewing angle for visible wavelengths. As the pixels become smaller, the viewing angle increases slightly: at 2 microns, 7.5 degrees, and 15 degrees for 1 micron pixels. Approximately.
Alright, it's time to head back to the lab.

You've just lost your MIT privileges for today: no more Doppler radar, hypertext, ethernet, fax machines, Bose speakers, GPS, disposable razors, RSA public key encryption, or spreadsheets until this time tomorrow.

Source: http://www.mitadmissions.org/topics/pulse/mit_inventions_breakthroughs/

On the contrary, most holograms that you're familiar with are 2D, recorded on either film or digital detectors. (Even volume holograms are only a few microns thick, for that matter.) I think it's also fair to note that holography recreates the original wavefront, which doesn't imply anything beyond a 2 1/2 dimension representation, enough for most people.
While the concept (ie, digital holography and spatial light modulators) isn't new, I'd like to think there's something with the processing chips which might be interesting and help fuel the current research and DH products.

Vagrant's right that it's about the light, but there's always more to the story... like that light gathering ability is proportional to the area and the sensitivity. Sony actually has a 56% advantage over OmniVision in terms of raw pixel size; of course, compared to PhaseOne's 9 micron CCD, that's 26 times less area. (Not to mention raw QE and sensitivity.)

The other key factor, at least in my mind, is the ability of the lens to make use of the pixels. A perfect lens focuses light to a spot that is limited by diffraction; for a nominal cell phone cam with NA=0.2 and green light, the spot is around 3 microns, twice the size of an OmniVision pixel. So even though the OmniVision has 8 MPx sampling ability, the imaging system has a throughput SBP of closer to 2 MPx.

That, of course, is why DARPA has the MONTAGE project...

The slipstick and sliderule have now been replaced by Java and Matlab. You know, something more useful than a slipstick for the average Course 6er, and it's got a better ring than "OOP via C++, MIT!" The good news is that the beaver cheer is still getting used.

More on topic, if you're looking to check out some hot beaver action, the 2.007 finals are online: http://web.mit.edu/webcast/2.007/2008/mit-2.007-finals-08may2008-220k.ram

Frank, this sounds like it is based around plenoptic cameras (or integral photography), which have been around for awhile. If you visit http://graphics.stanford.edu/papers/lfcamera/, they have a good video that shows some of the uses of plenoptics circa 2004-2005, such as the "3D" capture, rotating viewpoints, and digital refocusing.
So, if plenoptic cameras have been around for so long, why is this one making the news? According to the press release, it sounds like this is the first time that the imaging, sensor elements, and electronics have been integrated together onto a dedicated chip -- as opposed to modifying a current camera to do the plenoptic thing.
You might also check out coded aperture photography as another interesting way of encoding 3D information into a single photograph (http://groups.csail.mit.edu/graphics/CodedAperture/, also with cool video), or digital holography as a more quantitative research tool.

This method seems to have its greatest benefit in the large-format size and "simpler" display surface... not necessarily speed. For your holographic workstation, there's always Michael Bove at MIT (http://obm.media.mit.edu/) whose group is doing real-time holographic displays, and the group from U. Texas SW (http://innovation.swmed.edu/research/instrumentation/res_inst_dev3d.html) using gels for real-time holographics.
While we're never going to have Princess Leia (them photons don't just back-project for fun ya know), give it a few more years before one of these various 3D display techs takes off.

To be all technical, the ultra small radius/length ratio suggests that the fluid flow is going to be dominated by viscous forces -- precisely where Bernoulli fails. Beyond that, mass conservation suggests that a smaller cross sectional area will lead to higher average velocities, which is probably what you're thinking.

For comparison, the abstract quotes 150-300 um needle diameters; a (tiny) 33 gauge hypodermic needle has a 203 um OD...but only a 90 um ID bore (says Wikipedia). So these needles are comparable to other needles used for drug delivery. The benefit may be that they could use a thinner wall, achieving the same ID and same flow rates, but with a slightly smaller OD.

There, that was painless.

What, no ubiquitous SkyNet joke yet?