AMD Doubles Up on 8K
At the recent NAB event, AMD’s Xilinx business (acquired by the processor maker earlier this year) highlighted that its chip designs could process multiple uncompressed 8K streams. We thought that was interesting, so we spoke to Rob Green, Senior Manager – Pro AV, Broadcast & Consumer at AMD Xilinx, to get more background.
Xilinx has long had a position in ProAV and broadcast for its FPGAs (Field Programmable Gate Arrays). These chips tend to be used to accelerate performance beyond what you can do with software or discrete circuits. But when you do not have the volume or revenues to pay the high fixed cost of a dedicated chip or application-specific IC (ASIC) design, that can be a more cost-effective long-term solution. ProAV and broadcast markets tend to be relatively low volume markets but demand high performance, so they tend to be markets that are good for the FPGA technology.
Take codecs, for example. This market is highly asymmetric. Advanced codecs take a considerable amount of processing to do the encoding with limited volumes. Conversely, decoders have to be in every set and have potential markets in the millions making them better candidates for ASICs. Further, while there are usually straightforward ways to design and test a decoder, encoding often has many variables, and know-how and experience are essential in getting the best out of the process. That may mean changes in the circuits used as experience is gained in using the codec. Changes are a potentially massive cost in an ASIC, so FPGAs make a good compromise in early markets.
Having said that, Green told us that its FPGAs are used in some TCONs (the timing controller built into LCD and OLED panels), and as a result, the company’s technology is inside at least 18 current consumer TVs. The technology is also used in driver chips for LEDs of different kinds.
The firm’s chips are in many high-end cameras, including cinema devices, as the 112 Gbit bandwidth can handle that level of data, even with high bit depths. The firm has standard cores to handle mezzanine codecs such as IntoPix Tico and JPEG-XS, used in production environments.
Green told us that the chips could handle two streams of 8KP60 at one time for transcoding or processing using the firm’s Versal technology, which is based on a 7nm process node (its lower cost Ultrascale technology is based on a 16nm node). He added that the announcement of the 8K capability had also attracted those who wanted to process even more UltraHD and 4K video channels!
Versal includes support for Xilinx’s ACAP architecture, including other types of dedicated chip hardware, such as CPUs and GPUs in the chip. The architecture allows processes to be handled by the optimum kind of processing.
As well as video-related technology, Xilinx has a developed business in chips used for connection to PCI, PCIe, and PCI-X for use in data centers. Again, in this application, specialized hardware vendors may not have the volume scale to go to ASICs. However, the knowledge Xilinx has built in this area also includes networking. With the move of video away from SDI and HDMI towards AV over IP architectures, it has found itself in an excellent position to support clients looking to make that move. The company’s technology can support 100GB ethernet, which is getting more common and affordable.
We asked about any changes since the AMD acquisition, but Green told us that it is “very early days,” and although there look to be some clear synergies in what AMD has done in heterogeneous computing, little has happened yet at the technology and product level.