The pursuit of efficiency in cloud computing has led industries to innovate resource consumption, with Amazon EC2 Spot Instances emerging as a cost-effective choice for high-performance computing (HPC) and Electronic Design Automation (EDA) as more customers adopt Cadence Cloud. Amazon web services (AWS) EC2 Spot Instances offer a gateway for users to utilize unused EC2 capacity at significantly reduced costs, sometimes up to 90% less than standard on-demand rates. Despite these cost benefits, Spot Instances come with several caveats, such as using Spot Instances for workloads that involve prolonged intervals. As EDA workloads grow in complexity, effectively utilizing AWS EC2 spot instances becomes challenging, especially for workloads that demand prolonged execution. For instance, digital implementation tasks involving Cadence Innovus may take a few days. Such issues have deterred many from capitalizing on these cost-effective computing resources.

EDA Workloads Spot instance Support For Cost Efficiency

Nonetheless, EDA design is undergoing a seismic shift as collaboration between industry giants unlocks unparalleled cost-efficiency and reliability for users. Cadence has teamed up with MemVerge and Amazon Web Services (AWS) in an innovative alliance to take advantage of the robust EC2 Spot Instances, even for prolonged EDA workloads. This innovative initiative promises to redefine the approach toward managing long-duration tasks by merging advanced computing capabilities with economical solutions. MemVerge offers a system capable of capturing the state of an active task, archiving it, and resuming the operation on a different device.

Innovating Beyond Interruptions

Amazon EC2 Spot Instances are spare compute capacity in the AWS Cloud available at steep discounts compared to On-Demand Instance prices. The only difference between an On-Demand Instance and a Spot Instance is that a Spot Instance can be interrupted by Amazon EC2 with two minutes of notification when EC2 needs the capacity back. Such interrupt warnings can be detected within the running instance or externally via Amazon EventBridge. Yet, this adaptive approach was not feasible for EDA and numerous other HPC applications due to the lack of in-built interruption handling capabilities—until now.

It is essential to transfer these prolonged high memory jobs between machines to take advantage of Spot instances before the interruption of Spot Instance. EDA customers benefit significantly as they typically expend extensive compute resources daily. This convergence of efforts among Cadence, MemVerge, and AWS marks a significant leap forward, leveraging the advantages of Spot Instances for an expanded suite of EDA and HPC tools, all while keeping current applications intact.

This joint effort offers a technique for saving the state of processes and local temporary files to shared storage, effectively making applications interruption-proof. When a Spot Instance is halted, the checkpoint data can be leveraged to resume processes seamlessly on a new instance, thus maintaining the integrity of prolonged tasks without losing progress or data. So now we can use Spot Instances for long-running jobs like handling larger designs with Innovus, and this is crucial for cost saving because a customer, rather than allocating a single expensive machine, could allocate several different spot instances over that period of multiple days and reduce their cost without compromising the performance using the MemVerge infrastructure.

Cadence Innovus Implementation System Enhanced with AWS and MemVerge

Utilizing the incremental checkpoint technology developed by MemVerge, even high memory jobs like Innovus can proceed without disruption. Incremental checkpoints entail saving the process at periodic intervals and reducing the volume of data transferred during Spot interruptions. While this approach does increase the total data written, as changes accumulate over each checkpoint, it offers a strategic tradeoff by ensuring less data loss in critical scenarios. Key considerations include the recovery cost associated with infrequent checkpoints, the duration of each checkpoint, memory usage, and job length. Frequent checkpoints could save significant work in long-duration jobs but necessitate balancing time efficiency and data protection. This enables the completion of elaborate and extended compute jobs confidently, mitigating concerns about spot interruptions erasing vital advancements.

Spot Instance Benefits for Cadence Users

Significant Cost Savings: Using Spot Instances equates to massive cost reductions, freeing up budget for other critical areas or scaling up operations.

Enhanced Flexibility and Unmatched Reliability: MemVerge revolutionizes HPC by offering a combination of enhanced reliability and flexibility without prohibitive costs. Its advanced checkpointing solution safeguards computational projects against interruptions, which is particularly beneficial in volatile Spot Instance environments. This allows ambitious, compute-intensive tasks to be pursued confidently, mitigating fears of data loss or project failure.

Streamlined HPC Tasks: This streamlined, user-friendly experience abstracts the underlying complexity, making elastic compute more accessible to a broader audience.
Cadence, MemVerge, and AWS collaborated to demonstrate the Innovus implementation run moving between different AWS spot instances. The test design power, performance, and area (PPA) results were comparable to a traditional single run on an identical AWS compute instance, showing the MemVerge snapshot process did not impact the design quality of results.

Innovus Implementation System

At innovative technology nodes, designers face the challenge of balancing power, performance, and area (PPA) with the speed of design completion (TAT). Emerging physical and electrical complexities, alongside novel structures like FinFETs, demand no compromises on PPA or TAT to stay ahead in the competitive landscape. The Cadence Innovus Implementation System stands out by offering a solution that eliminates the need to compromise. Designed for dense layouts at advanced and established process nodes, it offers a significant PPA improvement of 10%-20% and accelerates TAT by up to tenfold. Featuring the industry's premier massively parallel approach, it efficiently manages extensive blocks, scaling up to 5-10 million instances. The system enhances capabilities in critical areas such as placement, optimization, routing, and clock management. Its innovative architecture considers the entire design flow, reducing iteration and speeding up the process. By leveraging the Innovus Implementation System, you gain the tools to develop highly integrated and unique systems with minimized risk.
Through their combined efforts, Cadence, MemVerge, and AWS are setting a new bar for HPC tasks, ensuring that businesses can confidently leverage the power of elastic compute, backed by the assurances of cost-effectiveness and continual operation.

MemVerge and AWS will jointly present this solution and test results at CadenceLIVE Silicon Valley on April 17, 2024, at the Santa Clara Convention Center! Register NOW.

For more insight into how EC2 Spot Instances can benefit long-running, delve into the existing AWS blog post "Save up to 90% using EC2 Spot - even for long-running HPC jobs".

Learn more about the Cadence Cloud portfolio.