Xilinx seems to finally be serious about AI/Vision at the edge
With the announcement of the first Kria SOM, Xilinx has at long last decided that edge AI/Vision is a market. In my former company, we could not contemplate any XCUV platforms because of the completely out-of-bounds cost of those chips.
The XCUV5, on which the KRIA 260 is based, still lists for $1500 -$2500! My former company would have loved to use it as a platform for its smart surveillance cameras but the price was too high by a factor of 10. The capability of H.264/H.265 compression is key for any platform offering video. Apparently they thought they could milk the broadcast video market with this offering - but that pricing cut the heart out of the (larger) video surveillance market entirely.
In the 2010-2018 timeframe. the capacity to run video analytics in real time was not available except in proprietary solutions (my former company had one such solution). Perhaps this is what prevented Xilinx from entering the market.
Cut to 2021, with various open-source neural-net object recognition frameworks. Now Xilinx has seen the writing on the wall with Google, Nvidia/Jetson and Intel/Movidius offering much lower cost platforms for AI/video. So now they have responded with the KRIA SOM, which is a fantastic step forward at $250/$350. Apparently competition is good!