> Cost of entering oPoW hardware manufacturing will be much lower due to Silicon Photonics using old CMOS nodes (~90 or 220 nm vs. ~7 nm for transistors).
If silicon photonics using old process nodes is so competitive, cheap to manufacture and power-efficient, why aren't they planning to use it for general purpose workloads? Matrix-vector multiplication is a key operation for most GPU compute, including for machine learning. They say "...On the other hand, to be competitive in the field of machine learning, one needs to compute billions of multiply-accumulate operations per second" but that kind of scale out applies just as much to crypto mining, and is the part that's supposedly addressed by reusing old process nodes.
If silicon photonics using old process nodes is so competitive, cheap to manufacture and power-efficient, why aren't they planning to use it for general purpose workloads? Matrix-vector multiplication is a key operation for most GPU compute, including for machine learning. They say "...On the other hand, to be competitive in the field of machine learning, one needs to compute billions of multiply-accumulate operations per second" but that kind of scale out applies just as much to crypto mining, and is the part that's supposedly addressed by reusing old process nodes.