AKIDA benchmarking?, page-105

Your enquiries are perfectly reasonable.

Thanks Sparkster. The weird thing is that honestly I am as keen as anyone for the company to succeed. I will open a bottle of champagne when, finally, the people at BrainChip provide the numbers that demonstrate that AKIDA can do problems that other systems can do (like NVIDIA Jetson) but either (a) much faster, (b) using much less power, or preferably both. If it turns out that BrainChip has been sat on such information for several months when it could have just published the numbers and convinced far more people than they could with a dozen fancy powerpoint shows or podcasts, then what should we conclude?

By the way, people keep asking why I am obsessed with frame rates. Here's one reason. Frame rate is very close to reaction time. When NVIDIA reports that its Jetson AGX Xavier can do 1919 frames per second on the SSD Mobilenet-V1 dataset (which has 300*300 pixels) https://developer.nvidia.com/embedded/jetson-benchmarks, what that means is that the reaction time to an image of a tiger (or whatever) is going to be about 0.5 msec. That is very fast, and fast enough to avoid the tiger! And if you are in autonomous vehicles, those milliseconds count. But the Jetson AGX Xavier is a whopping big chip that uses 10Watts or more. And costs $899 even if you buy 1000 of them https://developer.nvidia.com/embedded/faq#jetson-devkit-not-for-production

BrainChip's AKIDA could quite easily be a much better choice for many situations. Add to that the fact that it can also do native spike based processing and real time onchip learning with JAST, and it is potentially a real game changer.

I honestly believe it. But until BrainChip actually decides to publish some hard numbers, my judgment is more like a religious belief based purely on faith. Much like many people on this forum.....