"can model up to 17 million parameters" So about about 1000x away from being useful? Most current models are billions not millions.
Neuromorphic hardware is more about action potentials per unit time than raw parameter count.
IBM analog AI chip could give the Nvidia H100 a run for its money
11–20 of 71 posts
Re: IBM analog AI chip could give the Nvidia H100 a run for its money
#12Did they buy mythic maybe
https://www.theregister.com/2022/11/09/mythic_analog_ai_chip...
Re: IBM analog AI chip could give the Nvidia H100 a run for its money
#13Re: IBM analog AI chip could give the Nvidia H100 a run for its money
#14"can model up to 17 million parameters" So about about 1000x away from being useful? Most current models are billions not millions.
Neuromorphic hardware is more about action potentials per unit time than raw parameter count.
But unless it can directly or indirectly support current transformer architectures (with their huge parameter counts), it's going to critically struggle with adoption.
Re: IBM analog AI chip could give the Nvidia H100 a run for its money
#15And how about implementing this directly on the camera sensor? Before the A/D conversion.
Re: IBM analog AI chip could give the Nvidia H100 a run for its money
#16Earlier quoted context omitted.
Neuromorphic hardware is more about action potentials per unit time than raw parameter count.
But unless it can directly or indirectly support current transformer architectures (with their huge parameter counts), it's going to critically struggle with adoption.
Why can't we develop different architectures?
Re: IBM analog AI chip could give the Nvidia H100 a run for its money
#17Re: IBM analog AI chip could give the Nvidia H100 a run for its money
#18And how about implementing this directly on the camera sensor? Before the A/D conversion.
Arguably that is what the Foveon sensor did.
https://en.m.wikipedia.org/wiki/Foveon_X3_sensor
There is a book called The Silicon Eye about the history, and not coincidentally Carver Mead also wrote a book on neuromorphic analog neural network implementation back in the 80s.
Re: IBM analog AI chip could give the Nvidia H100 a run for its money
#19Re: IBM analog AI chip could give the Nvidia H100 a run for its money
#20And how about implementing this directly on the camera sensor? Before the A/D conversion.
Arguably that is what the Foveon sensor did. https://en.m.wikipedia.org/wiki/Foveon_X3_sensor There is a book called The Silicon Eye about the history, and not coincidentally Carver Mead also wrote a book on neuromorphic analog neural network implementation back in the 80s.
Hmm all I've seen about Foveon is just stacked RGB sensing. Ie there is no computation involving multiple pixels or anything like that.