PORTLAND, Ore. -- Freescale Semiconductor's embedded DSP targets scalable video coding (SVC) using a customizable H.264/SVC algorithm for its StarCore DSP. The chip maker said its design enables cost-effective media gateways, high-end video conferencing systems and IPTV distribution systems by eliminating the need for Pentium-grade processors for coding.
"Today you need something like a fast Pentium processor to do SVC, making it both cost and power-consumption prohibitive for media gateways and the end points in most video conferencing systems," said John Dixon, media marketing manager at Freescale. "What we now offer with a DSP-based solution is a much lower price point for use with gateways, as well as the flexibility to scale down in the future to even smaller, less expensive DSPs-based devices."
SVC enables a single encoder to create a video streams that contains several sub-bitstreams that can be separately decoded by dropping packets, allowing algorithms to down-sample for lower spatial resolution, slower frame rates and lower quality--or a combination of the three--as required for specific viewing devices. Hence, it is said to solve problems with stutters and stops, choppy frames, delays, blurred motion and broken pictures suffered by traditional video conferencing systems.
The MSC825x DSP is the first to implement the H.264/SVC algorithm. It is touted a fully programmable part that can handle realtime 1080p video streams at 30 frames per second, but which can be down-sampled by clients to as slow as 176p at 7.5 frames per second. The StarCore-based DSP also encodes and decodes H.264/SVC, and handles over 30 other legacy voice and video compression algorithms.
Dixon claimed Freescale MSC825x is twice as fast as the Texas Instrument C64x. The MSC825x also comes with software migration tools for translating existing TI DSP code into Freescale DSP code, for one-, two-, four- or six-core processors.
Adam Segal, are you saying a programmable DSP programmed to compresses video is cheaper to implement than a custom dedicated ASIC that is designed to compress video for a specific application? Please. On what planet? Your lead paragraph talks about replacing Pentiums. Hardly a competitive statement that elicits notions of cost effectiveness. If you are saying there are no SVC processor custom ASICs(arguable) so the user has no choice other than a DSP, that's hardly proving your point of cost or market superiority. It just shows the market isn’t big enough to justify a custom ASIC. A DSP has its place. Flexibility, uniquely customizable, openness to smaller volume opportunities and time to market should be on your slides, not low cost and high volume. As volumes grow the DSP becomes a victim of the flexibility that won it the socket in the first place.
Programmable DSPs are able to perform many functions due to the nature of the programmable DSP core processor and not with the aid of specific HW blocks. Hence, if you are using the DSP cores for Video you are not paying any overhead and majority of customers will utilize all or most of the device components for a typical video application
In addition, most video applications enabled by this SW solution will require little or no additional chips or components beyond DDR memory.
With regards to ASIC solutions, there is no fully featured ASIC solutions for HD SVC available today in the market neither for encoding nor decoding
Lastly, I’d like to touch on programmable DSP suitability for high volume applications as well as low volume ones. DSP devices offered by Freescale are shipping and planned to ship in tens of thousands per customer for high volume cost sensitive applications. The inherent advantages of such devices make them an ideal choice for a variety of applications at any volume.
These programmable solutions are nice for low volume, expensive applications. The problem is a programmable DSP is designed to do many things, so the overhead one pays for things one does not use is very high. Further, these DSP vendors leave a lot of peripherals to be added externally, further adding to the cost. If one wants to compress video inexpensively get a custom ASIC that compresses video. Questions like, how much memory is required, the cost of the support chips required for an app and the cost of the silicon in volume is conveniently absent for this press release.
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.