It's genuinely 100% meaningless until it is compared against other products using the same benchmarking frameworks and use case scenarios.
I hope someone comes up with standard benchmarks that could be applied to these AI accelerators and neuromorphic processors. Similar to the software used for CPU or GPU benchmarking, like Sysmark or 3DMark, in both contrived and real-world use cases.
While we've already seen NVISO post some numbers for their specific API use cases in behavioral analysis, I'd like to see Akida, both 1.0 and 2.0 processors creating training data sets from sensor input.
While I imagine most users will focus on the chip's inferencing abilities, I would love to see how a small cluster of Akida processors would perform instead of using expensive GPUs in the cloud to create the models themselves.
its a big fat
