Advanced Computing in the Age of AI | Thursday, March 28, 2024

Six Nines IT & Astera Labs – Case Study: Revolutionizing Semiconductor EDA Workflows with AWS 
Sponsored Content by Six Nines

Introduction / Executive Summary

In Electronic Design Automation (EDA), limits on storage, memory, or processing capacity can impact product quality, project budget and delivery schedule. However, on the public cloud, customers can scale up resources without any disruption, bursting from literally dozens of cores on a master control node to thousands of cores within minutes with the ability to scale out and utilize the compute power of the cloud. This case study describes how Astera Labs successfully performed the first large-scale chip design from start to finish on the public cloud.

There are many benefits of moving HPC manufacturing workload processing to the public cloud as it provides virtually limitless scalability and flexibility for varied applications and allows customers to scale compute up and down very quickly. Running in the cloud saves investing in depreciating capital equipment that is often underutilized and allows staff to focus on their area of expertise rather than IT infrastructure and code optimization.

Compute intensive workloads from HPC simulations, artificial intelligence (AI), machine learning (ML), deep learning (DL) and Internet of Things (IoT) generate massive amounts of data. Organizations are increasingly moving AI, ML and DL and manufacturing processing/storage to the cloud rather than on-premise data centers. Hyperion Research indicates that total HPC market spending (including public cloud) grew from $22 billion in 2013 to $26 billion in 2017 and is projected to reach $44 billion in 2022 1].

Astera Labs Selects Six Nines to Create their AWS Cloud Environment

Astera Labs Inc. (Santa Clara, CA) is a cloud-based semiconductor start-up specializing in purpose-built connectivity solutions for intelligent systems. Their products include system-aware semiconductor integrated circuits, boards and services to enable robust PCIe connectivity. Astera Labs started with a clean slate. They had no on-premise data center but expected to grow rapidly and wanted a system to grow with them. Their expertise is in chip design and not IT, so Astera decided to operate entirely in the public cloud because managing an on-premise data center and hybrid cloud requires managing two solutions.

Astera selected Six Nines as the partner to architect and configure a complex cloud environment using Amazon Web Services (AWS) and Synopsys for their EDA solution. Using the Six Nines’ system, Astera Labs created the first large-scale chip design fully implemented and verified from start to finish on a third-party public cloud using AWS. They developed the industry's first PCI Express® (PCIe®) 5.0 retimer for heterogeneous compute and workload-optimized servers.

“We chose to go with the public cloud because it’s extremely scalable and elastic so we can go up and down very quickly, and we can outsource the support. We certainly didn’t know how to get started on AWS, but we were lucky to partner with Six Nines and they were able to set this up for us.”
Jitendra Mohan, CEO, Astera Labs

Six Nines AWS Solution for Astera Labs

One of the only Premier-level (AWS) Consulting Partners focused on HPC, Six Nines has extensive experience helping customers run compute intensive EDA, Computer Aided Engineering (CAE) and AI/ML workloads. Six Nines combined AWS-specific expertise with extensive cloud EDA experience to help Astera greatly accelerate their time-to-market by architecting, configuring and maintaining their end-to-end environment. They worked closely with Astera to implement best-in-class tools for creating a design and testing workflow in AWS, including Synopsys EDA for simulations.

The Amazon NICE Engine Frame web portal allows Astera to manage users and have a comprehensive view of all machines and jobs that are running across the compute fleet. Six Nines incorporated the latest instance types to help Astera leverage the nearly infinite resources and scalability of AWS, including Amazon Elastic Cloud Compute (EC2) C5, R5 and Z1D instances, all powered by Intel. Amazon EC2 R4 was used as the front end for the storage node, which is tuned to be a shared storage node for all workloads. In order to realize the full elasticity of running their simulations in the cloud, Six Nines configured the environment so Astera can utilize the cloud resources as close to 100 percent as possible and then turn them off when they’re finished.

Figure 1. Astera Labs system created by Six Nines

Cloud CIO™ Software Provides Cost Optimization and Billing Tracking

Six Nines provided Astera with cloud cost and usage data via Cloud CIO software. “One of the unknowns when using the cloud is how much cost savings you can expect, and how to optimize it. CloudCIO is a service that provides Astera Labs with operational oversight for cost management monitoring, billing and policy management,” states Mitch Becker, HPC Specialist at Six Nines.

Figure 2. Benefits of using Cloud CIO and Amazon EC2

Six Nines Solution Drives Product Innovation for Astera Labs

Six Nines successfully designed and deployed a complex end-to-end chip-design environment, entirely in AWS, which helped Astera accelerate their business model and reduce time-to-market. Because AWS and Six Nines enabled easy and affordable access to thousands of computing cores and storage, Astera's engineers were able to add more features to the chip without compromising quality, budget or delivery schedule.

Traditionally with EDA, no amount of storage is enough. On AWS, Astera can scale up storage without any disruption, bursting from literally 32 cores on a master control node to thousands of cores within minutes with the ability to scale out and utilize the compute power of AWS. “When we ran into an issue, we could increase our storage 2x or 4x without any disruption,” says Mohan. “It literally took us five minutes to go from 99% usage back down to 50%. And that was really beneficial.”

By designing an independent compute infrastructure so the verification and physical design instances could run in parallel, Astera was able to add a few features to their chip that they would not have been able to do otherwise. “With this solution, we were able to develop a much higher quality chip in a compressed schedule to exceed expectations from our customers. When we needed to ramp up from some level of resources to 2x or 3x higher, that happened overnight, literally in two or three days. Putting together that infrastructure, that quickly, on-premises would have been nearly impossible, not to mention the costs associated with it,” adds Mohan.


References

1] HYPERION RESEARCH UPDATE. Presented at the 2019 ISC19 Conference, Slide 30.

 

About Six Nines

 

Remove guesswork and trial and error with Six Nines IT

“By now, given the significant opportunities related to flexibility and cost savings, many of the largest and most profitable manufacturing companies have long-since migrated HPC workloads to the cloud. Now, the rest of the industry is working to catch up. However, cloud migrations and running workloads is full of challenges and risks that can offset the key benefits of a migration. And you can no doubt imagine that there are many important considerations—both during your initial migration as well as once you’re established in the cloud.” Adds Becker Even if you’re confident in the ability of your team to manage the transition, if they are inexperienced in working in the cloud, then it will likely take significant trial and error to dial in your systems and processes.

If you’re looking for a way to remove guesswork, reduce risks and streamline the transition, Six Nines IT can help. Our cloud experts understand that every migration and workload implementation is unique and that taking a cookie-cutter approach can only deliver limited value. We’ve performed hundreds of cloud transformations over the last decade and have culminated our extensive experience to develop a five-phase Cloud Adoption Framework (CAF). Designed to help accelerate customers’ cloud journey while reducing risk, this framework ensures the architecture, planning and design of each environment align with the five best practices pillars in the AWS Well-Architected Framework, which include:

  • Operational Excellence,
  • Security,
  • Reliability,
  • Performance,
  • Cost Optimization.

In other words, once we review customers’ goals, network topology and recommend a reference architecture, you can rest assured you will be on a fast, responsible, and highly cost-effective path to cloud success.

Learn more

For more information, please, or contact us at: [email protected], (415) 937-7070, or visit Contact Six Nines  Copyright 2019

 

EnterpriseAI