By Sean Thielen
For any established semiconductor product developer, designing a next-generation PCIe 5.0 chipset in less than a year is no small feat. For a brand-new startup with no compute infrastructure other than laptops, however, it is a huge ask. That’s why, with time being of the essence, Astera Labs decided to take a chance on the efficiencies it would gain from a 100% cloud-based approach. Astera Labs estimates it reduced the development time for its initial semiconductor chip offering by 30-40% compared to using a traditional model while still ensuring a high-quality product that meets the stringent needs of their demanding customers.
Removing Bottlenecks in Compute-intensive Workloads
Headquartered in Silicon Valley, Astera Labs is a fabless semiconductor company that develops purpose-built connectivity solutions for data-centric systems. The products in Astera Lab’s portfolio, which include system-aware semiconductor integrated circuits, boards and services to enable robust PCIe connectivity, are designed to eliminate performance bottlenecks and integrity challenges in hyperscale environments. “Astera Labs has partnered with industry leaders to tackle challenges related to the mainstreaming of AI and other cloud-based compute-intensive workloads. In that context, there are a variety of new challenges from a system implementation standpoint,” explained Jitendra Mohan, CEO of Astera Labs. “What Astera brings to the table are the unifying connectivity components that are essential for connecting all of the processing elements on the server motherboard and seamlessly enabling them to communicate with each other.”
Building a Cloud-based Semiconductor Company
The team at Astera Labs did not have a plan to become a cloud-based company from the beginning. According to Mohan, when the company began considering its infrastructure options in late 2017, it was focused on maximizing the long-term value. “We expected to grow quickly, so we were looking for a solution that could evolve with us. After looking at on-prem, hybrid, electronic design automation (EDA) vendor and cloud options, the public cloud made the most sense for the way we work and we opted to go with Amazon Web Services (AWS),” he explained.
The decision to pursue a cloud-based model, however, was not without trepidation since no one on the Astera Labs’ team was a cloud expert. “Our strengths are in chip design and not in IT and managing EDA and we didn’t know how to get started on AWS,” noted Mohan. At the same time, the team was eager to start developing new products and running simulations, so it reached out to Six Nines IT, a Premier AWS Consulting Partner, for help. Six Nines not only brought deep AWS expertise to the table, but it was also familiar with EDA workloads. Jason Cutrer, founder and president of Six Nines IT, noted that Six Nines is one of the only AWS consulting partners focused on HPC in the cloud. “There are a lot of premier AWS consulting partners that are strong public cloud consulting partners, but we’re really the only one that is focused on HPC,” said Cutrer. “We also have a lot of experience with EDA-specific workloads.” Mohan said his team was running simulations roughly two weeks after engaging Six Nines.
Based on recommendations from Six Nines, Astera Labs relies on a variety of Intel-powered AWS instance types. For simulations, it primarily uses Amazon EC2 C5, R5 and z1d instances. For general purpose tasks, it mostly relies on EC2 R5, R4, z1d and Spot instances, among others. Mohan noted that gaining immediate access to the z1d instances once they became available was a significant benefit of being in the cloud: “The z1d includes new custom Intel® Xeon® Scalable processors, which are proven to be fast in the cloud and ideal for EDA workloads, so we were eager to get access to them. The option of running many jobs in parallel on the z1d instances really sped up our simulations.”
Seeing Advantages on Multiple Levels
According to Sanjay Gajendra, the chief business officer of Astera Labs, the company’s cloud-based development approach has enabled the company to move and scale at a speed that the company’s customers need and expect. “When we started out, we were hesitant to take this path not knowing what we were getting into,” said Gajendra. “But looking back on what we gained, meaning the amount of productivity improvement we achieved, simplistically we saved significant development time compared to the same set of people trying to do the same thing using a more restrictive compute approach.”
Gajendra said that using the cloud was also game-changing from a quality assurance standpoint, since Astera Labs could rely on independent compute instances for physical design and verification. “We would never have been able to do the same number of simulations and checks in a traditional environment… In the cloud we could fire 1,000 or 1,500 cores and run verification jobs in parallel with physical design to test the heck out of our chip in a relatively short time. It meant we were able to move very quickly while also ensuring a quality product.”
Growing in the Cloud and Getting Better
Although the team at Astera Labs is now confident that its cloud-based compute approach was the right decision, it acknowledges there is still a lot to learn and significant room for improvement. “We are always looking at how to allocate dollars to get the greatest returns. With our initial project, things weren’t always optimal because we were learning as we went about how the machines we use and other factors in job flows impacted costs,” Gajendra explained. “As we move into new projects, I can clearly see some opportunities that will translate into greater savings going forward.”
Read other case studies about HPC in the Cloud:
- Gompute’s HPC in the Cloud Delivers Performance
- Penguin Computing On Demand Redefines HPC in the Cloud
Intel technologies’ features and benefits depend on system configuration and may require enabled hardware, software or service activation. Performance varies depending on system configuration. No computer system can be absolutely secure. Check with your system manufacturer or retailer or learn more at www.intel.com.
Cost reduction scenarios described are intended as examples of how a given Intel- based product, in the specified circumstances and configurations, may affect future costs and provide cost savings. Circumstances will vary. Intel does not guarantee any costs or cost reduction.
Intel does not control or audit third-party data. You should review this content, consult other sources, and confirm whether referenced data are accurate.
Intel, the Intel logo and other Intel marks are trademarks of Intel Corporation or its subsidiaries in the U.S. and/or other countries.
Other names and brands may be claimed as the property of others
© Intel Corporation
Speak Your Mind