Executive Summary
Astera Labs, Inc., a fabless semiconductor company headquartered in California’s Silicon Valley, develops purpose-built connectivity solutions for data-centric systems. The company’s product portfolio includes system-aware semiconductor integrated circuits, boards, and services to enable robust PCIe* connectivity. The company went from being a brand-new startup to having a fully designed next-generation PCIe 5.0 chipset in less than a year using a novel cloud-only business model running on Amazon Web Services (AWS). As an AWS user, Astera Labs was among the first semiconductor companies to take advantage of the custom Intel® Xeon® Scalable processor in the Amazon EC2 z1d* instances. By relying on the cloud for all of its high performance computing (HPC) needs, 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.
Challenge
Astera Labs, with a passionate vision to be “the trusted partners to distribute data in intelligent systems,” started with a completely clean slate in late 2017. Founded by industry veterans, the company had a strong chip design background, but it lacked expertise in running IT systems and managing electronic design automation (EDA). After carefully evaluating the tradeoffs between on-premises, hybrid, or cloud-based IT, Astera Labs chose a 100% public cloud approach based on options for outsourcing support and priorities around business flexibility and compute elasticity among other factors.
According to Jitendra Mohan, CEO of Astera Labs, although his team was eager to get moving with the cloud-based approach it didn’t know how to get started. “Astera reached out to AWS, who in turn referred us to Six Nines IT, a Premier AWS Consulting Partner,” said Mohan. “Six Nines explained our different options on AWS and based on their advice we did our own study and put together our preferred approach. Initially, we weren’t making any effort to set up the most efficient file server possible and those sorts of things. We just wanted to be able to run simulations like we were used to and get moving on our designs. It only took a couple of weeks from having literally nothing other than our laptops to being able to run some simulations.”
Solution
Jason Cutrer, founder and president of Six Nines IT, said that Six Nines helped Astera Labs architect, configure, and maintain their end-to-end chip design environment, which is 100% in AWS. “One of the advantages on partnering with Six Nines is that we have a lot of industry experience with EDA-specific and HPC workloads, so we were able to greatly accelerate their configuration and time-to-market with an infrastructure to assist in their design and simulations,” explained Cutrer. “As we engaged, we made recommendations to use a variety of Intel-powered AWS instance types, which helped accelerate Astera’s timelines.”
As part of its initial efforts in the cloud, Astera Labs worked on designing a new semiconductor chip. “For simulations, we predominantly used Amazon EC2 C5 and R5 instances along with EC2 z1d instances once they became available. For the backend, we rely on different types of machines, including the z1d, R5, and R4 instances. We also use other instance types for general purpose tasks,” stated Mohan.
When we put together our initial architecture, the z1d instances weren’t even released yet. Once they became available, we used them heavily for simulations and backend as well.” — Jitendra Mohan, CEO of Astera Labs
Depending on size, the C5 instances include either custom 2nd Generation Intel Xeon Scalable processors, 2nd generation Intel Xeon Scalable Processors or the 1st generation Intel Xeon Platinum 8000 series processors. The instances with custom 2nd generation Intel Xeon Scalable processors have a sustained all-core turbo frequency of 3.6 GHz and maximum single core turbo frequency of 3.9 GHz, making them ideal workhorses for simulations. The others can achieve a sustained all core Turbo frequency of up to 3.4 GHz, and single core turbo up to 3.5 GHz using Intel® Turbo Boost Technology.
With custom Intel Xeon Scalable processors, the z1d instances are among the fastest in the cloud. They provide high single thread performance and sustained core frequency of up to 4.0 GHz. Even higher performance is possible with the help of Intel® Turbo Boost Technology, which accelerates processor and graphics performance for peak loads when cores are operating below specific power, current and temperature thresholds. “When we put together our initial architecture, the z1d instances weren’t even released yet. Once they became available, we used them heavily for simulations and backend as well,” said Mohan.
Jitendra Mohan, CEO of Astera Labs, explains the company uses the latest HPC instances on AWS to provide execution speed and reliability.
Result
The z1d instances are specifically designed for the EDA industry, so being able to start using them as soon as they were available in the public cloud was a huge benefit. “As the saying goes, ‘time is money’ and we started using the z1d because of processing speed. The availability of z1d instances and the opportunity to run many jobs in parallel gave us the ability to run jobs much quicker than the older instances and that directly contributed to our success,” explained Mohan.
Mohan noted that the performance and reliability of the Intel® processors in the various AWS instances have met and exceeded expectations since starting in the cloud: “With processors, execution speed is very important, which is why we chose the C5 and z1d instances, but reliability is also critical. Obviously, a glitch in the middle of a job that needs to run for three to four days is really problematic. In the year or so of our first design project, we experienced thousands of machines that were rock solid from an infrastructure standpoint with almost no crashes.”
Overall, the benefits of the cloud approach included faster turnaround times and faster iterations on designs and simulations. “We were able to create a higher quality chip much faster than would have been possible using our own resources. Doing what we did on-premises would have been incredibly expensive and probably impossible.”
Solution Summary
By operating exclusively in the cloud, Astera Labs enjoys a high degree of operational flexibility. It not only can easily scale AWS compute resources up and down as needed to keep design processes speeding along and costs in check, it can also employ the resources that make the most sense for a given workload. AWS offers compute instances that include several generations of Intel processors and technology, including custom Intel Xeon Scalable processors and 2nd generation Intel Xeon Scalable processors. The ability to immediately take advantage of the latest Intel technology as new AWS compute instances become available will be an advantage for Astera Labs moving forward.
Technology Highlights
- Amazon EC2 z1d instances—custom Intel® Xeon® Scalable processor
- Amazon EC2 C5 instances—2nd Generation Intel® Xeon® Scalable processor
- Amazon EC2 R5 instances—Intel® Xeon® Platinum 8000 series processors