Discipline and Process Built Emerald Rapids

Behind the Builders: Irma Esmer Papazian works on the performance and power for the upcoming 5th Gen Intel Xeon processor.

Talking to Irma Esmer Papazian, senior principal engineer, data center processor architecture, is a lesson in perspective. Growing up in Istanbul, Turkey, she had never touched a computer when she chose her career path: “Computer engineering was a field that was in demand – it was at the top of the list for the college application process, so I figured it must be interesting and likely is the future.”

That one decision, driven by a curiosity of the unknown, marked the beginning of a decadeslong career in computer engineering, including 30 years working in Intel’s data center group. It’s a busy place to be.

This week, Intel launches 5th Gen Intel® Xeon® processors, code-named Emerald Rapids, which deliver performance and efficiency gains and increased cache, and arrive only four fast quarters after the debut of 4th Gen Intel® Xeon® processors (Sapphire Rapids). Working on performance and power with the Data Center and AI Group, Esmer Papazian helped to launch both.

4th and 5th Gen Intel Xeons Bookend 2023

Starting and ending the year with a Xeon launch is quite a feat. Esmer Papazian credits precision planning, updated workflows and a one-team approach for the tight turnaround, though it’s clear a double-digit performance/power improvement wouldn’t have been possible without expert engineering informed by years of experience with prior generations of Xeons. If you’re looking for Xeon expertise at Intel, Esmer Papazian is about as experienced as they come.

“I grew up at Intel,” she says, adding that she’s spent more than half of her life here. She came to the U.S. for graduate school with the intention of getting her Ph.D. But in 1992, after completing her master’s degree, Intel recruiters came to campus and changed her mind.

“It was a dream job,” Esmer Papazian says of the role, where she was the youngest person in her group and the only woman. She learned a ton from her senior colleagues. She then spent almost a decade on Itanium before shifting to Xeon in 2005. Suffice to say, in the 18 years since, she’s seen it all. That was crucial for Emerald Rapids, which required a team that blended new and old faces to rethink the execution process.

As Esmer Papazian and the Sapphire Rapids team worked to get 4th Gen Intel Xeon out the door, work had already begun on the next generation of Xeon. Irma joined Emerald Rapids at tape-out, just as 4th Gen Xeon wrapped. While much of the team shifted to Granite Rapids (due in 2024), Irma jumped into her second Xeon launch inside 12 months, bringing her 4th Gen Xeon experience to the new group.

Giving Customers What They Want

The directive with Emerald Rapids was simple: Customers were thirsty for more performance, but – as usual – didn’t want a hefty increase in power consumption that would drive up costs.

A change in the chip topology helped move things in the right direction. While 4th Gen Intel Xeon had up to four chiplets, Emerald Rapids has two larger dies with increased density that allows for more cores. Irma says these larger clusters with increased L3 cache capacity, four channels of memory and 32 functional cores are better suited to large virtual machines. They also benefit from reduced latency and improved power consumption.

With silicon set, Esmer Papazian and the team began figuring out how they could increase performance and improve power efficiency even further post-silicon.

“We internalized that feedback. We said, ‘Okay let’s look, what are our options without any silicon changes?’”

The team identified a list of features they could deliver against that feedback at the end of an A-0 silicon execution. In the end, there were roughly a dozen features and optimizations that made the grade. This included improvements on existing energy efficiency features (an enhanced active idle mode feature extends the prior generation’s active idle mode detection via core activity checks and includes changes to the threshold for exiting idle mode), as well as new features not present in 4th Gen Xeon: a fifth turbo level that allows workloads with high AVX512 utilizations to reach higher frequencies.

As they drove the new feature definitions and tuning, they also looked at what hadn’t gone so well on Sapphire Rapids, to make sure they could execute at the pace necessary to stay on time. Debug templates and analysis guidelines were developed to help newer members get up to speed and drive consistency. They held regular status meetings and showed their work often.

“We tried really hard to capture and share the analysis flows and methods in depth to train and develop the entire team,” Esmer Papazian says.

The team created dashboards that they used to push updates up and across, rather than focusing inward and only pulling data when someone came knocking. The entire team was focused on automation and streamlining the processes. For a group of engineers, it was a lot of process and tracking, which Irma says was the hardest part of the journey.

“As technologists, we want to work on technical problems and solutions,” she says. “We want to spend less time on program management and processes. But the timelines of the program and our desire to deliver above our original targets meant we had to invest in process.”

The team also spent extra effort ensuring future generations of Xeon could benefit from it all.

Esmer Papazian explains: “Any issue we encountered has a one-pager. High level: What was the problem, the root cause, how did we fix it and what should we do on next generations to not run into the same thing? I am super proud of the entire Emerald Rapids team. Validation, architecture, design and manufacturing all came together to deliver the best possible performance on this product within the targeted schedule.”

Leading with Compassion – or How to Solve Problems that Don’t Exist Yet

Esmer Papazian applies the same forward thinking in day-to-day collaboration, where she says people are busy all the time and it’s easy to get bogged down and hyper-focused on the task at hand.

“At some point, you are working so hard that you lose the perspective of the big picture. I try to emphasize the big picture, how what one is doing relates to the final goal. Setting perspective helps. Also, I remind people that we’re not alone. We are a team, and we should benefit from the collective knowledge and expertise. Don’t try to re-solve a problem, see if it is already solved, and improve,” she explains, adding that giving credit where credit is due goes a long way to helping people feel they are valued members of the team and that their contributions matter.

“If somebody is doing the work and putting the effort in, they should be feeling that they are represented well and getting visibility for the work they are doing. As we work as a team, we emphasize that.”

Esmer Papazian believes that encouraging people to ask for help, recognizing their hard work and fostering a culture that respects people’s personal needs is what keeps the present in perspective and ensures future problems are shared and halved rather than shouldered alone in isolation.

“You have to remember, as engineers, we are constantly inventing and pushing the boundaries of technology. This is not the last problem that I’m going to solve, I’m going to keep having them and I’ll be tapping on the same resources.”

That’s especially true in this cutting-edge technical environment. Experts are harder to come by and their time is in high demand. With Emerald Rapids wrapped and ready for launch, she’s turned her attention to something that is on everyone’s mind: artificial intelligence.

“Right now, I’m looking at AI and large language model performance scaling on CPUs,” she says. “There’s always a fresh challenge. New problems to solve.”

Intel will launch 5th Gen Intel Xeon on Dec. 14 in New York City at its AI Everywhere event.