At Macworld last month, Ambric announced that it is using its massively parallel processor architecture as the basis for PC plug-in video accelerator boards. The new video platform is based on Ambric’s AM2045 programmable processor chip and includes off-the-shelf video codec software written by video codec house MainConcept (which was acquired by DivX late last year). OEMs can buy a PCI Express-based reference board from Ambric and add their own codecs or pre-/post-processing software, then sell the board to end users. End users, in turn, get a plug-and-play video accelerator that’s integrated with Adobe’s Premiere Pro and After Effects video authoring tools; a check-box option in the Adobe GUI directs the applications to offload video processing to the Ambric board.
Initially, Ambric will include MPEG-2 and H.264 encoders along with ten different decoders, all running under Windows XP. Support for the MAC OS X Leopard operating system is planned for the second quarter of 2008.
Ambric expects that the video accelerator boards from OEMs will cost in the range of $2,500 to $3,500 depending on the number of encoders included, and consume about 15 watts of power. (Two OEMs, Sorenson Media, and Pyre AV by ADS Tech, have already announced products based on Ambric’s platform.) Much as digital audio processing made its way down from the pro realm to anyone with a computer on their desk, desktop video processing—including encoding and transcoding, not just playback—is slowly filtering down to more users. But video is much more computationally demanding than audio. High-definition H.264 encoding (used in HD DVD and Blu-Ray discs, for example) is particularly challenging, and requires a powerful processing engine to achieve anything approaching real-time performance.
Ambric’s massively parallel chip architecture is based on a large array of RISC processors connected in a MIMD (multiple instruction, multiple-data) fashion. The AM2045 chip has 336 processors running at 300 or 350 MHz. Initially the motion estimation component of the MPEG-2 and H.264 encoders will be offloaded to the Ambric chip, along with a couple of other (undisclosed) H.264 tasks. Ambric plans to offload more functions later. The company claims that its board can accelerate H.264 High Profile, Level 4 encoding by a factor of 2-5X relative to a software-only implementation on a 3 GHz quad-core CPU. This isn’t enough for real-time encoding, but is a considerable speed-up.
Founded in 2003, Ambric is one of a slew of new massively parallel chip vendors hoping to find a niche for their chips. Such vendors all face the same problem: system designers are hungry for the extra processing horsepower but leery of complex chips, unfamiliar programming paradigms, and the low survival rates of start-up vendors. By providing a complete application solution rather than just throwing a chip over the wall, Ambric expects to alleviate some of these concerns and accelerate its chip’s acceptance in the market. And of course, they’re hoping to jumpstart their revenue stream while they’re at it.
It’s a sound strategy, and Ambric isn’t the first massively parallel chip vendor to try this approach. picoChip, for example, offers WiMax (and more recently, femtocell) reference designs, and has managed to get their chips lodged into WiMax products from multiple vendors. While picoChip is focused almost exclusively on a few wireless applications, Ambric intends to use the video accelerator cards to get a leg up and then branch out into a wider range of applications. From Ambric’s perspective, once the chip gets past customers’ resistance to new architectures and programming paradigms (both of which are hidden in the new product) the architecture will be a proven solution, which just might enable the Ambric architecture to take off.
Add new comment