Crafting and Testing the Do-Anything Server Chip

Sometimes the way to improve the performance of something is not to give it more power or capabilities, but rather to give it less work. Let it focus on what it does best.

Instead of having your brilliant head chef wash and chop vegetables, you hire help and let her spend more time crafting fine dishes. You get higher quality and more creative plates. And you get more of them.

The analogy plays out in the newest 4th Gen Intel® Xeon® processor, explains Nevine Nassif, an Intel Senior Fellow and Xeon chief engineer. In its case, the head chef is the microprocessor core and the kitchen assistants are a collection of specialized accelerators available on the chip.

"You don't want to waste the time of these multipurpose, very powerful cores doing simple compression crypto algorithms or moving data around," Nassif says. "Accelerators do that work much faster and more efficiently - and save the core to do more important work."

The result is a leap in performance efficiency. When accelerators are part of the work, the 4th Gen Xeon processor (code-named Sapphire Rapids), introduced in January, delivers on average almost 3 times more performance per watt than its predecessor¹.

Chief Task of the Chief Engineer: Converting Ideas into Silicon

The big gains were years in the making. In the unique role as chief engineer, Nassif saw them through from beginning to end.

After the chief architect sketches out the features of a chip with input and expertise from marketing and a range of technical specialists, the chief engineer teams up to figure out if and how those features can be built, tested and delivered.

It's not unlike when the executive chef or hotelier asks the head chef to put a new twist on a bouillabaisse stew on the menu - the ask might be clear enough, but the head chef needs to figure out the actual ingredients and procedures to deliver a large volume of perfect bowls, day in and day out.

Early on, the tasks are translating the architecture to the actual design with the talent, tools and time available. As the product comes to life, the role becomes more chief problem-solver, dot-connector and listener.

"I tend to be a very optimistic kind of person," Nassif says, "so it's very important to make sure I have enough people around who will give me the bad news and who will bring me back down to Earth." That also means being open to new ideas, which can come from anywhere.

"There's a certain number of things I know really well - there's a lot I don't. I depend on others to help me through it, and there are some very good people out there."

Nassif turns to many a sous chef - experts in specific domains of the chip, in fabrication and assembly, in testing and validation and many others across the company - over the course of development.

'Path-Clearing' to Plug-and-Play Chips

It's been well documented that the Sapphire Rapids project was particularly challenging. The product introduces several first-ever technologies and manufacturing techniques for Intel: new memory, new interfaces, new accelerators. And it's the first time a Xeon chip was built by combining multiple die on a single package (referred to before as "making chips like quilts").

"The dream is to do a lot of plug and play," Nassif says, where a chip comprises a colorful array of individually tuned chiplets. New chips could be introduced more quickly (and with more variety) by mixing selected new/updated chiplets with already-proven, off-the-shelf chiplets for routine functions.

"Sapphire is path-clearing the technology to get you there," explains Nassif. Close to the plug-and-play dream, the 4th Gen Xeon package combines four identical chiplets to amass a new level of total capability. In other words, it's employing all possible options for higher performance: not just new kitchen assistants, but more head chefs, too.

That path-clearing meant figuring out not only how to make it, but also how to test it. Which is further complicated by the relentless expansion of the modern data center. "Our customers are constantly surprising us in the way in which they stress our parts in ways we haven't thought of," Nassif says.

A typical Xeon processor installed in a big cloud data center might barely ever rest, handling a nonstop and dramatic variety of applications.

Finding Old Bugs - and New Limits

As customers continue to push even decade-old products in new ways, they occasionally find bugs - from a simple typo to electrical interference between two unrelated components - that might also exist in newer processors. "Any time we hear something back, we go back and look," Nassif says. "Why did we miss it? What can we do differently?"

Teams will run new tests across generations of the product, Nassif explains, and existing tests will be examined and improved at every step in the product's lifecycle. "You want to find bugs as early as possible," she notes. "So we're always trying to learn."

And with projects running in parallel - Nassif is also chief engineer for Emerald Rapids, the successor to Sapphire Rapids - lessons are applied immediately.

"Engineering is very creative," Nassif says. "A whole bunch of it is art."

With every new challenge overcome and bug resolved, the art becomes science as processes, tools and people improve. But they know the next challenge is never far behind.

"You have to be open to trying things and to failing," Nassif says. "There's no magic - just ideas, and you keep at it."

About Nevine: Builder in Brief

Home Site: Hudson, Massachusetts

Title: Intel Senior Fellow and Xeon chief engineer

Team: Xeon Engineering Group

Path to Fellow-ship: How does one become the chief engineer of the most important product at the company? "I was very determined to see it through - and I did."

Perfect Weekend Day: "Sleep in as late as I can. Go swimming for an hour or so. Have a nice lunch and then read a book - a murder mystery. At the end of the day, maybe do some gardening."

Sources of Inspiration: "Listening to others and being open. You have to find the co-travelers, the people that will come with you and don't think you're totally crazy to do this."

¹ _{See [E1] at intel.com/processorclaims: 4th Gen Intel® Xeon® Scalable processors. Results may vary.}