AWS for Industries

Using Cadence’s Pegasus Physical Verification with TrueCloud, customers benefit from 2X walltime savings using Amazon EC2 X2iezn Instances

Introduction

Silicon designers are creating increasingly complex designs to address ever-increasing demand from customers. Increased complexity poses challenges to electronic design automation (EDA) applications that help with manufacturability and functionality of the designs on silicon. Physical verification is an integral part of the silicon design system that identifies design layout issues prior to its production on silicon. Physical verification is comprised of many flows that include, but are not limited to, design rule checking (DRC), layout versus schematic (LVS), and more.

To address today’s design landscape, Amazon Web Services (AWS) collaborated with Cadence, a leader in electronic design with more than 30 years of computational software expertise that applies its underlying Intelligent System Design strategy to turn design concepts into reality.

This blog post describes their test using Amazon Elastic Compute Cloud (Amazon EC2) X2iezn Instances, and the Cadence Pegasus Physical Verification with TrueCloud. Cadence successfully executed physical verification jobs on Amazon EC2 X2iezn Instances and achieved up to a 50 percent performance improvement over Amazon EC2 R5 Instances.

Physical Verification Compute Requirements

Historically, physical verification jobs have been very compute- and memory-intensive, requiring a large share of CPUs and memory. Silicon designers are often resource-challenged to run physical verification on designs that can consume thousands of CPU cores and require multiple days to complete. The Cadence Pegasus Verification System, a cloud-ready, massively parallel physical verification tool, is designed to overcome these issues. The new Pegasus Physical Verification with TrueCloud helps designers to run physical verification jobs from on-premises compute resources onto the cloud without having to copy any design layout, schematic, or foundry rule decks to the cloud. It has a massively scalable architecture that can reduce design cycle time significantly by using the thousands of CPUs offered by AWS.

AWS introduced Amazon EC2 X2iezn Instances to address multifaceted EDA workload requirements. It also recognized that customers performing physical verification require increased clock speeds and a large memory footprint. Amazon EC2 X2iezn Instances are powered by a second-generation Intel Xeon Scalable processor, which offers up to 4.5GHz, the highest frequency in the cloud. They feature up to 1.5 TB of memory and deliver up to twice the performance per vCPU than X1e Instances. Amazon EC2 X2iezn Instances offer 32 GiB memory per vCPU and provide up to 48 vCPUs and 1536 GiB RAM. Amazon EC2 X2iezn Instances are built on the AWS Nitro System, delivering up to 100 Gbps of networking bandwidth and 19 Gbps of dedicated EBS bandwidth. Amazon EC2 X2iezn Instances are built for workloads that require high performance per thread and a high memory footprint of up to 1.5 TB.

Pegasus Physical Verification with TrueCloud on Amazon EC2 X2iezn Instances

The Pegasus system’s cloud-based infrastructure allows execution on the cloud, enabling silicon designers to avoid a CPU bottleneck during the physical verification execution cycle. Using the AWS cloud, a silicon designer using the Pegasus system can easily scale compute resources up or down as needed. This model, known as Pegasus Elastic Compute, delivers an efficient method that is very well suited for running jobs on the cloud.

The Pegasus system’s architecture allows it to achieve near-linear scalability using thousands of CPUs. Traditionally, designers would have had to use a complex system to run physical verification on the cloud. They would transfer their designs, schematics, and foundry decks into the cloud followed by several setup files. The Pegasus Physical Verification with TrueCloud eliminates all of that complexity. Compute operations are done on the cloud while the IP (design and foundry) stay on premises. Designers have only to upload the Pegasus software to the cloud and begin testing, which is a significantly faster way to operate.

The Cadence and AWS teams worked together to evaluate the performance of Pegasus workloads on AWS and configure infrastructure using Amazon EC2 X2iezn Instances.

Performance testing using Pegasus Physical Verification with TrueCloud on AWS

Testing of Pegasus performance on AWS was performed using Amazon EC2 X2iezn Instances. Cadence used an on-premises compute system with eight CPUs, which read in the design layout, schematic, and foundry decks. Using that data, the Pegasus system launched compute jobs on the cloud at the desired scale.

Typically, on-premises machines tend to be of older generation and slow down physical verification runs. Taking advantage of the latest instances, like Amazon EC2 X2iezn Instances, which have the latest processor and large memory, speeds time to tapeout. Cadence achieved a 38 percent performance gain in a cloud run that used the Pegasus Physical Verification with TrueCloud with respect to an on-premises run that used the same number of cores (240).

Cadence also did a comparison with another design to compare turnaround times on premises versus using Pegasus Physical Verification with TrueCloud versus using a traditional cloud method where all the data was transferred to the cloud and then run on the cloud.

For this design, Cadence observed a 40 percent turnaround time improvement for Pegasus Physical Verification with TrueCloud versus on-premises. Also, there was no significant difference between Pegasus Physical Verification with TrueCloud versus a traditional cloud run (rows 2 and 3 in table 2), which proves the efficiency of using this model.

Finally, Cadence did another run comparing Amazon EC2 R5 Instances versus Amazon EC2 X2iezn Instances (table 3).

For this design, Cadence observed a 50 percent turnaround time improvement running on Amazon EC2 X2iezn Instances versus on Amazon EC2 R5 Instances.

Conclusion

Using Amazon EC2 X2iezn Instances to run the Cadence Pegasus Physical Verification with TrueCloud to test a physical verification EDA workload showed significant walltime savings, which will drive cost and time savings. Customers who run EDA workloads can now use the new Amazon EC2 X2iezn Instances from AWS to obtain an up to 50 percent turnaround improvement versus the Amazon EC2 R5 Instances using the Pegasus system.

Amazon EC2 X2ezn Instances offer 40 percent performance improvements versus on-premises systems when the CPU count is the same and are excellent for running memory-intensive physical verification workloads.

For further information, contact an AWS sales representative.

Click to learn more about the Pegasus system.

For more information about Cadence, visit www.cadence.com, and for more information about EDA workloads on AWS, visit https://aws.amazon.com/semiconductor.

TAGS: , , , ,
Ahmed Elzeftawi

Ahmed Elzeftawi

Ahmed Elzeftawi is a Semiconductor and EDA Partner Solutions Architect at Amazon Web Services. Prior to joining AWS, he was a Cloud Product Management Director at Cadence Design Systems where he architected and deployed multiple compute clusters for EDA workloads on AWS. He has 20 years of experience in Electronic Design Automation, chip design, and High Performance Compute clusters. Ahmed holds a Bachelor of Science in electrical and communications engineering from Cairo University and an MBA from Santa Clara University.

Art Baudo

Art Baudo

Art Baudo is a Principal Specialist, EC2 Core, WWSO (Worldwide Specialist Organization). Art has over 20 years of experience in the tech industry, where he has worked in software architecture, product management, product marketing, CPU and GPU roadmap/architecture, and technical marketing. Art has held technical and managerial roles at Motorola, Intel and AWS. He has a BS in Electrical and Computer Engineering from Rutgers University and a Masters in Information Technology from Northwestern University.

Dibyendu Goswami

Dibyendu Goswami

Dibyendu Goswami is a product engineering architect at Cadence who is responsible for the development, qualification and deployment of Pegasus physical verification tools and flows. He joined Cadence in 2014. Prior to that, he worked for Texas Instruments and Intel in standard cell library development, floorplanning, full-chip integration, tapeout and related areas.