Not your Great Grandfather’s Switch
Surprising as it may seem, your great Grandad would understand the principle behind just about all network switch architectures on the market today. Namely, the crossbar, used in early telegraph and telephony switches. While such designs are acceptable for small slow systems, they do not work well for interconnecting clusters. Efficiency is reduced to single digits so your processors are consuming lots of energy just waiting for data.
Why? Almost all existing switches are only 50% efficient, are congestion prone, do not scale and have high latencies.
It is time for a change and we have gone back to first principles to develop a switch that can scale to hundreds of ports, still maintain a low latency and has high fault tolerance. They will at least double cluster efficiency.
Our first switch will be CXL 3.1 compatible and can scale to hundreds of 1000Gb/s ports while maintaining a 50ns average latency.
Further details coming shortly