中文 English

Astera Labs: Purpose-Built Connectivity

Startup targets compute-intensive workloads in the data center.

popularity

Growing amounts of data are forcing companies to rethink where data is processed and when, how and where it is moved. But solutions may vary greatly from one company to the next, and from one use case or application to the next. This is forcing the adoption of a heterogenous compute architecture that combines traditional processors, such as CPUs, GPUs and FPGAs, with AI processors and smart network interface cards on the same motherboard or within the same rack in a data center.

“What we’re seeing is that the traditional server architecture no longer cuts it,” said Sanjay Gajendra, chief business officer at Astera Labs. “We’re heading toward composable disaggregation, which is more of a distributed model. So rather than dedicated banks of memory and storage, we’re looking at a virtual system with X amount of memory, and that is provisioned when and where the workload is processed. This is the next paradigm shift for the data center or the edge.”

The challenge is adding that kind of flexibility into an architecture. Most data centers are set up to scale up to the necessary processing using fixed memory and processor architecture. So a cloud architecture can add more resources as needed, and turn them off when they are not required. But utilizing those resources more flexibly requires a different way of looking at the problem.

“We’re building technology to remove connectivity bottlenecks and distribute data in intelligent systems,” said Gajendra. “Processors within a heterogenous compute system can generate and consume lots of data, this requires fast and low-latency interconnects. Our Aries Smart Retimers for PCIe enable 32 gigabit/lane transactions compared to today’s 8 gigabit/lane – this 4x jump in performance is a critical enabler.”

Gajendra noted that Astera Labs, which operates entirely in the cloud, has been working with EDA vendors to improve performance of their tools. “EDA tools were designed for local servers, not thousands of parallel threads,” he said. “That was the biggest challenge. We worked closely with the EDA vendors to implement a cloud-based EDA flow to accelerate time to market, reduce costs, and deliver a high quality product”

Astera Labs was founded in 2017 by CEO Jitendra Mohan, along with Gajendra and Casey Morrrison, who heads systems and applications. The company is currently in series B funding from Sutter Hill Ventures, Intel Capital, Avigdor Willenz Investment Group, GlobalLink1 Capital and VentureTech Alliance.

Astera Labs’ Aries Smart Retimer portfolio is now in production for PCIe 4.0 solutions, with sampling underway for PCIe 5.0. The company also launched a cloud-scale interop program with ecosystem partners in support of customers deploying its product in their cloud systems.



Leave a Reply


(Note: This name will be displayed publicly)