.svg)
Late last year I wrote about how NVIDIA had made hardware sexy again, after software had eaten the world a decade earlier. So last week at Cisco Live in Amsterdam, it was very pleasing to see, not only many old colleagues and friends, but also the old Cisco driving hardware innovation.
For me the biggest announcement was the new Cisco Silicon One G300 AI chip. This chip can provide 102.4 Tb Ethernet switching and brings Cisco in-line with the two other fastest Ethernet chips on the market – Broadcom’s Tomahawk 6 and NVIDIA’s Spectrum 6. For anyone interested in the maths behind 102.4Tb, the G300 chip provides the following.
- 200Gb Integrated SerDes (Serializer/Deserializer), which is the speed you can get on any one lane.
- 512 Ports High Radix Scaling, which is the number of lanes that you can run 200Gb on.
- 200Gb * 512 = 102.4Tb of total switching capability.
- This will provide 64 * 1.6 Tb Ports, each of which is made up of 8 * 200Gb lanes.
Cisco also announced platforms powered by the G300 chip, including new Nexus N9364F-SG3 switches to allow scale up in the data centre, and Cisco 8000 series routers to scale AI workloads across data centres.
I was also excited to see the Cisco Nexus N9324C-SE1U and N9348Y2C6D‑SE1U smart switches, which contain the AMD Pensando DPU. These have price parity with other Cisco Nexus ToR switches and could see a change in how we provision network policy within the data centre. More to follow on this in coming weeks.