The results from the performance presentation were absolutely amazing. Akida was able to exceed 1000 fps (over 1600 fps in the best case) while the other two (Jetson Nano, and Tegra X1 GPU) did not even pass that threshold. Still yet, it did this at a 1/3 of the clock cycles of the Jetson, while roughly 1/5 of the cycles for the Tegra. It would be interesting to see the power drawn by each for the testing.
I really like the slide in the presentation that shows the transition over time. They've classified BrainChip as "Extreme Edge". It should be pretty clear at this point that BrainChip's goal is to license their IP, and allowing their customers to incorporate it into their own edge solutions. They are taping out with different manufacturing processes as a proof of concept to demonstrate the flexibility of their technology.
A certain individual (who shall not be named) will attempt to convince people that companies could choose any competitor to BrainChip as long as doing so ultimately gives them a working solution. While logically this is true, the statement would indicate to me a lack of formal training in engineering, where the idea is to not only provide a solution, but doing so in a way that requires less power, provides superior performance, and most importantly, keeps costs low. Choosing the technology that can meet both those criteria, while having an extremely strong showing against its competitors in benchmarks makes that argument seem rather silly.
I would think that one working on an edge-based solution that can benefit from AI inference, after looking at the slide below (as well as that of the other benchmarks from NVISO's presentation), it certainly gives the reader an impression as to the best choice of technology for their solutions. It would certainly warrant a closer look and taking time to evaluate the technology.
View attachment 29331