Broadcom brought some interesting technology to the table at AI Infrastructure Field Day 3. We got to physically put our hands on the Networking Chips being used in AI Servers..

Not being heavily involved in the AI side in the data center, it was interesting to see the size of these chips.
At AI Infrastructure Field Day 3, Broadcom started the day off discussing how AI Infrastructure is changing from relying heavily on Infiniband and switching over to standard Ethernet, albeit an upgraded version than what is in you home and office connecting your Wi-Fi APs and switches, called Ultra Ethernet.
One of the biggest chatter in the tech space is about how much power and bandwidth AI is needing. Broadcom said during their presentation that Bandwidth growth is insane but we need more even though it sounds like we have enough we when start talking the speeds of these chipsets.
We can throw an infinite amount of bandwidth at AI, it will be used instantly.
That is why AI Infrastructure Field day exists because all the talk is about AI changing the world, but how are we connecting those servers doing both the training and the inferencing?
Enter the Tomahawk..
Tomahawk 6

Broadcom’s Tomahawk 6 chip was physically showcased to the delegates. It contains 1000 serialize/deserialize lanes of throughput of which 100-gigabit PAM-4 signaling to achieve a total of 102.4 Tbps. Combining the two optical engines each pushing 51.2TB. It is a AI Chip with an Ultra Ethernet switch built right in to push all the data as fast as possible to where it is needed when it is needed. The designs of these data centers are insane.
When it comes to these Ultra Ethernet, shorter passive copper cables => better SNR, less power requirements.

The Tomahawk 4 is just optimizing AI world UPS or Fedex delivery services to get your packets there faster. AI is pushing the needs for how much data can be pushed through a cable leading us back to the age old quote:
Never underestimate the bandwidth of a station wagon full of tapes hurtling down the highway”
Is that saying being left in the dust? What is coming next in this AI data thirsty world? Where will we go next to meet the demands of AI? Is this sustainable? Only time will tell.
Check out the whole Broadcom event at AI Infrastructure Field Day 3 here.

