The Cloud Advantage: Optimizing PPA and Delivery with Cadence Cerebrus

Graphics processing units (GPUs) have significantly transcended their original purpose, now at the heart of myriad high-performance computing applications. GPUs accelerate processes in fields ranging from artificial intelligence (AI) and machine learning to video editing and complex scientific simulations. The surge in mobile gaming, beginning around 2018, has particularly underscored the importance of efficient, low-power GPU architectures. This is an arena where Imagination Technologies continues to excel, consistently paving the way forward with innovative graphics processing solutions tailored for the burgeoning demands of both mobile gaming enthusiasts and data-heavy computational tasks.

 Transforming PPA and accelerating chip delivery while not wasting time or resources on setting up the environment or installing all the licenses have been the significant challenges for Imagination technologies. Further, we need more GPU pipeline stages to deal with high-resolution demands, which increases the power requirements. Further, with the reducing form factors, dealing with leakage power is another challenge. In their 5nm GPU design, Imagination used Cadence Generative AI-based Cadence Cerebrus Intelligent Chip Explorer and the Software-as-a-service (SaaS) environment where everything is run in the Cadence cloud. Cadence OnCloud platform is straightforward to set up and offers good results; it enables pay-for-use flexibility and massive scalability; it allowed Imagination to use Cerebrus on Cadence Cloud to improve the PPA of designs without any on-premises IT resources/setup requirements. To investigate the benefits of generative AI-based Cadence Cerebrus Intelligent Chip Explorer Imagination selected, multiple blocks in their recent 5nm GPU ranging in size from 500K to 4.3M cells. Cerebrus enabled Imagination to optimize its latest 5nm GPU automatically.

Our low-power 5nm GPUs had an aggressive delivery schedule and the AI-driven Cadence Cerebrus using a SaaS model gave us rapid access to a tailored Cadence-managed cloud environment on AWS—all without any CAPEX investment and the long wait times associated with on-premises hardware procurement and IT setup,” said Tim Mamtora, Chief of Innovation and Engineering at Imagination Technologies. “With the powerful combination of Cadence Cerebrus and the complete Cadence digital full flow, our team was able to automate processes that were previously manual, which saved us time and helped us achieve our PPA goals.”

Using AI-driven Cadence Cerebrus in a SaaS model provided the Imagination design team with the following benefits:

• Faster turnaround time and cost benefits- AI-managed peak hours and no additional resource requirements.
• Rapid access to massively parallel compute.
• Flexibility- engineering teams across the globe could access it from anywhere.
• Quick ramp-up with an EDA-optimized, ready-to-use, multi-cloud-enabled SaaS model that is security-compliant.
• Pre-trained Cadence Cerebrus AI models available to Imagination and Cadence customers as a warm start for their design projects.
• Improved productivity, enabling the team to achieve power, performance, and area (PPA) goals faster and accelerate time to tapeout.

Results

Imagination selected three 5nm blocks from the C-Series GPU IP, ranging in size from 500K to 4.3M cells. Cerebrus used the current Design Optimization Kit reference flow as a starting point, then optimized the blocks with timing closure as the primary objective and power and area reduction as secondary goals. As part of Cerebrus AI-driven design optimization, a machine learning model is generated, which can transfer training data between Cerebrus runs, significantly reducing run-time for further optimization. These pre-trained Cerebrus AI models are available to Imagination and Cadence customers as a warm start for their Imagination GPU design projects. Cerebrus used 20 distributed computers per block, each with 16 CPUs, so the peak CPU usage was 3x20x16= 960 CPUs.

Using the Cadence AI-driven digital full flow in the cloud, Imagination reduced leakage power on its latest 5nm by 20% and total power by 6% while improving area and performance.

“The AI-driven Cadence Cerebrus and the broader Cadence digital full flow are specifically designed for complex, next-generation designs like Imagination’s low-power 5nm GPUs,” said Dr. Venkat Thanvantri, corporate vice president of AI R&D in the Digital and Signoff Group at Cadence. “Through our collaboration with Imagination, they successfully leveraged AI-driven automation and the flexibility of our SaaS model to dramatically improve PPA and deliver a high-quality GPU design faster.”

Conclusion

Cerebrus can optimize compute and human effort, and this new SaaS ability puts it in a completely different league. The broader digital full flow provides optimal PPA and faster turnaround time, supporting Cadence’s Intelligent System Design strategy, which enables system-on-chip (SoC) design excellence. The Cadence Cerebrus, powered by generative AI, aims to help chip design teams become more productive and achieve significant PPA benefits. Its SaaS model allows teams to access nearly unlimited parallel computing and software license capacity, leading to faster turnaround times and cost-effective solutions. If you're looking for a way to increase design productivity, achieve higher PPA, and reduce time-to-market without the burden of significant Capex investment, costs, and delays of setting up and managing a secure IT environment, Cerebrus in Cadence-managed SaaS is the ideal solution for you.

Read More

• Imagination Optimizes PPA and Speeds the Delivery of Low-Power GPUs Using AI-Driven Cadence Cerebrus in the OnCloud Platform
• Cadence Cerebrus in SaaS and Imagination Technologies Case Study
• Cadence OnCloud
• Cadence Cerebrus Intelligent Chip Explorer