We are very proud and excited to introduce Innovium TERALYNX, a game-changing, new Ethernet switch silicon product family to power the best data center networks of the future!
Innovium’s TERALYNX is the industry’s first product line designed from the ground-up to be optimized for large data centers. The TERALYNX Ethernet switch family delivers options ranging from 3.2 to 12.8Tbps, the world’s highest performance and scalability with largest on-chip buffers, while at the same time featuring line-rate programmability, advanced analytics and deterministic low latency. Moreover, it delivers a 2x+ advantage in performance per watt, something that large data centers care about a lot.
Early customers and partners that we are engaged with have told us that they are excited with TERALYNX’s capabilities, and are working to introduce systems and other complementary solutions using our new chip and software. Some companies, mainly silicon providers, remain skeptical on how such a complete set of capabilities is technically feasible using 16nm process technology.
In fact, we’ve heard comments such as:
- “Do you guys have a very large die area?” (hint: it’s very similar in die area compared to other chips with half the bandwidth or lower)
- “What features did you cut to make this possible?” (hint: not anything that scale data centers care about)
- “Is it really just one die or many dies in one chip package?” (hint: customers don’t like multiple large dies in one package due to reliability, power and other concerns)
- “A switch chip is predominantly made up of SerDes and SRAM, so what is your secret sauce?” (hint: more on this below)
We began from a standing start in 2015 and wanted to focus on the data center, because it was clear then, as it is now, that IT is moving rapidly to the public and private cloud and traffic rates are increasing unabated. Many billions of people have yet to get access to social networking, video/image sharing and information delivered via large data centers.
What we noticed was that the vast majority of products deployed in data centers were originally created for service providers, enterprise or embedded applications. Not surprising because large data centers began to emerge only 6-7 years ago and they built their infrastructure using products that previously existed in the market. Other companies attempting newer silicon products deployed their engineers in areas which are not top priorities for data centers and made poor architecture tradeoffs resulting in higher power consumption, lower throughputs, worse latencies and smaller buffer sizes.
This presented a significant opportunity to use a very different approach. We spent time talking to key customers to understand capabilities that were critical and what features they didn’t really care for. More importantly, we spent time asking about their major pain points, strong desires and trend lines. We studied the ecosystem in terms of compute, storage and optics, to ensure everything would align nicely in the context of customers. Finally, we focused our team’s innovation and R&D efforts in top areas that truly matter to data centers.
The result has been extremely satisfying. We have accomplished groundbreaking and fundamental innovation that is embodied in a significant patent portfolio. Customer interest in TERALYNX is strong and we are delighted at the potential to make a lasting impact. By delivering an unmatched 128 ports of 100G in one chip, we are causing users to rethink network topologies and oversubscription ratios. TERALYNX analytics delivers the most advanced foundation for autonomous data center networks. Importantly, we believe TERALYNX will significantly accelerate the move to distributed micro-services, AI, increased storage traffic over Ethernet and enable exciting new video, IoT, 5G and AR/VR applications that the world hasn’t even dreamt of.
If you are building systems for large data centers, or operate large data centers, each of which have several hundreds to 10s of thousands of servers, we’d love to show how TERALYNX can help you differentiate. Please contact us at [email protected]
The Innovium Team