Optimizing Electronic Design Automation (EDA) Cost with the Intel® Xeon® Processor Scalable Family

Intel’s silicon products are increasingly complex, with literally billions of transistors on a single chip. As design complexity increases, so does the demand for more validation. This makes the design process increasingly time-consuming because each validation step takes longer to complete. Adding more compute power can help speed up the process; however, our data centers’ capacity and power envelope are not limitless. Therefore, we are deploying servers based on the 3rd Generation Intel® Xeon® processor Scalable family into our data centers. These innovative processors offer more compute capacity and performance than older processors, without increasing data center space and energy consumption. In addition, these processors help control EDA software licensing costs. Typically, we spend about $4 on software licensing for every $1 spent on servers. By using 3rd Gen Intel Xeon Scalable processors to maximize compute performance per core, we can meet our EDA workload requirements while reducing EDA application license growth rate.

Recent throughput and per-core performance tests comparing various Intel Xeon processor SKUs on EDA workloads have demonstrated that by upgrading to 3rd Gen Intel Xeon Scalable processors, we can significantly improve our capacity and compute performance. Based on the results of our tests, we are deploying servers based on these processors in our data centers. This deployment is also in step with our current refresh cycle. By doing so, we expect to significantly reduce time to market of Intel chips as well as reduce the rate of growing costs of EDA software licensing.

Choose the Right Processors for Increasingly Complex EDA Workloads

Silicon design encompasses three phases: front-end logic design, back-end physical design (including circuitry, compliances, power performance, etc.), and tape-in/tape-out processes required for chip manufacturing. Front-end and back-end designs afford the most opportunity for cost reduction, including licensing costs. While each new generation of Intel Xeon processors has consistently improved performance and throughput, our tests specifically identified which SKUs deliver optimum results within a given stage of design.

We ran front-end and back-end EDA tests comprising chipset design workloads using industry-leading EDA vendor tools that are single-threaded and multi-threaded.^[i] Our goal was to assess performance and throughput improvement by measuring the time taken to complete a specific number of design workloads. This enabled us to match the best 3rd Gen Intel Xeon SKU (core count and frequency) for a given EDA workload.

We ran tests comparing 8-core, 16-core, and two 24-core offerings from the 3rd Generation Intel® Xeon® Gold 6300 processor Series, measuring the per-core performance of each in a Register Transistor Level (RTL) Simulation. In this case, we found that the higher frequency Intel Xeon Gold 6334 processor delivered 1.26x higher RTL Simulation per-core performance for critical-path EDA runs, outpacing the per-core performance lower-frequency CPUs in this processor Series (see Figure 1).

Figure 1. Use a higher-frequency 3rd Generation Intel® Xeon® Scalable processor for better per-core performance.

However, when it comes to RTL Simulation throughput, a higher core count is what matters most. Our tests demonstrated that a higher-core-count CPU SKU, such as the Intel Xeon Gold 6342 processor, can deliver up to 2.56x higher RTL Simulation throughput per server when compared to a lower core-count CPU, which is ideal for volume validation runs (see Figure 2).

Figure 2. Use a higher-core-count 3rd Generation Intel® Xeon® Scalable processor for better throughput.

The take-home here is that it pays to invest in speed for performance-critical applications because the more quickly jobs complete, the lower the licensing costs for essential EDA software. But when it comes to throughput, a higher-core-count processor is needed to support cost-optimal computation.

Mind Your Refresh Cycle

We ran tests comparing EDA throughput across four generations of the Intel Xeon processor family. The test results showed that the latest 3rd Generation Intel® Xeon® Gold 6342 processor-based server provided an up to 2.76x increase in throughput per server compared to an Intel® Xeon® processor E5-2680 v4. In other words, four servers powered by the latest generation of Intel Xeon Gold processors are capable of doing the work of eleven older Intel Xeon processors E5-2680 v4-based servers. Fewer servers not only mean less data center space and energy cost but also lower licensing costs. These results underscore the value of timely server refresh. And if you’re using disaggregated servers, you can refresh the compute/memory module without having to replace other, still-viable hardware components—further reducing costs and e-waste.

In summary, increased design compute requires more EDA applications, which currently drives up the cost to approximately $4 for EDA licensing for every $1 spent on servers. Refreshing servers with technology that is tailored for the application is ideal to curb these costs and improve TCO.

For full test results and discussion, read the IT@Intel white paper, “Increasing EDA Performance and Throughput with the Intel® Xeon® Processor Scalable Family.”

________________

[i] Testing by Intel IT as of April 2021 through January 2022. 1st, 2nd, and 3rd Gen Intel® Xeon® Scalable processor tests were run on innovative disaggregated servers. For more information, see “IT@Intel: Green Computing at Scale.”