Sandra Riveras Next Mission: Do For FPGAs What She Did For Intel's Xeon

Sandra Rivera, a longtime Intel veteran and chief executive of the chipmaker's FPGA business, reprised the Altera brand during a webcast Thursday in which she shared her vision for the newly spun-off company.

Speaking to the press ahead of the event, Rivera painted a bright future ahead for Altera's FPGA business, predicting a total addressable market of $55 billion over the "next several years," up from $8-10 billion.

What exactly "several years" means isn't clear, but at a seven to eight percent compound annual growth rate, it's going be a while before we get there. However, the takeaway is clear, Rivera needs to assure folks, distracted by the massive gains driven by GPUs and other high-end ASICs for AI, that the FPGA market is alive and well.

This is because, even by Intel's own admission, Altera, which the x86 giant acquired in 2015, hasn't always had the best management, with CEO Pat Gelsinger saying as much when the chip biz announced the spinoff back in October.

To address this, Rivera aims to rebuild relationships with its channel sellers and expand Altera's portfolio to address a broader range of applications, particularly at the lower end of the spectrum.

"We have an expanding portfolio that allows us to deliver products and market from the very top in terms of that power, performance and capability, and features all the way to the most cost-sensitive, the most footprint-constrained and power-constrained environments," Rivera said.

The prioritization of high-end FPGAs at the expense of lower-end industrial products is seen as one of Intel's missteps in the years since acquiring Altera.

And Gelsinger, who was busy running VMware at the time of the acquisition, seems to agree, saying in October: "We also see that we have the opportunity to execute more effectively in the lower margin, mid and low-end areas of the business."

Bringing in the fixer

Gelsinger's decision to put Rivera in charge of executing on this vision shouldn't come as a surprise to anyone.

In fact, this isn't the first time that he's tapped the exec to clean up Intel's messes. In mid-2021, shortly after Gelsinger's return to Intel as chief executive, Rivera was put in charge of the Datacenter and AI (DCAI) group amid a series of executive shakeups.

In the role, she was responsible for both Intel's Xeon CPU family and FPGA businesses. However, it seems most of her attention centered on restoring Intel's ailing Xeon roadmap.

As you may recall, Intel's Sapphire Rapids Xeon scalable processors were due to launch at the end of 2021. The chip was among Intel's most ambitious to date, introducing several new features, including a chiplet architecture supporting higher core counts, PCIe 5.0, CXL, and support for DDR5 memory and high-bandwidth memory.

However, almost immediately after taking over the division, problems with the chip's development became evident. In a blog post published a little over a week after Rivera took over DCAI, Intel conceded the chip was behind schedule and would instead ship in the first quarter of 2022.

Except it didn't. As we later learned, Intel had bit off more than it could chew and after repeated delays, Sapphire Rapids arrived in January 2023.

But while Intel engineers were scrambling to get Sapphire Rapids out the door, the team was laying the groundwork for its next-gen Xeons. And by mid-2023, the chipmaker was already teasing the successors to its then-unlaunched Emerald Rapids platform.

The first of these, due out in the first half of this year, will deliver up to 288 efficiency cores based on Intel's 7nm manufacturing process — which has since been rebranded as "Intel 3" — and compete directly with AMD's 128-core Bergamo Epycs and Ampere's 192 Arm core AmpereOne processors.

Intel's Granite Rapids Xeons, meanwhile, are due out later in 2024 and will boast higher core counts, improved performance, and a substantial uplift in memory and I/O throughput, putting it in a better position to take on long-time rival AMD. You can find a more indepth run down these and other DC chips due out this year here.

But if getting Intel's Xeon processor family back on track weren't enough, Rivera was also thrown some curveballs along the way. One of the more notable came in late 2022 when Intel axed its Accelerated Computing Systems and Graphics Group (AXG). As part of the reorg, Intel's datacenter GPU and AI accelerator programs were shuffled under the DCAI umbrella, while Intel's gaming graphics platforms fell under the Client division.

Just a few months later, Intel revealed just how far behind its GPU roadmap was, canning its Rialto Bridge accelerators — the successor to the GPU Max chips used in the Aurora Supercomputer — and delaying its next-gen Falcon Shores platform, now just a GPU, to 2025.

Whether Rivera's leadership was what made the difference for Intel's Xeon division is up for debate, but she clearly made an impression.

A different kind of market

Winning share from rival AMD, which acquired FPGA market-leader Xilinx in 2020, won't be easy, analysts told The Register.

According to IDC analyst Brandon Hoff, AMD's Xilinx division leads the FPGA market by a wide margin, commanding about 55 percent share. Intel's Altera, meanwhile, holds the number two spot with control of more than 30 percent of the market. The remainder is held by Lattice Semiconductor, Microchip, and other smaller players.

If being the underdog weren't enough, Gartner analyst Alan Priestley told The Register that the FPGA market is unlike any other and is the polar opposite of Intel's core business.

At a high-level, FPGAs contain logic gates that can be reprogrammed on-demand to make the silicon behave differently. This makes them attractive in low-volume deployments where the economics of purpose-built ASICs doesn't make sense, and general-purpose silicon is either inefficient or under-performant.

According to Priestley, these qualities mean FPGAs tend to show up in two major segments: the first involving high-end FPGAs based on newer process tech designed to tackle complex, high-performance applications. Examples include things like smartNICS and 5G base stations. In these environments, performance and flexibility are prized over cost. Thus, FPGAs built for these applications have tended to be very low-volume and high margin.

The second category is far larger, encompassing industrial and embedded applications, like IoT and robotics. Here again, the flexibility of FPGAs remains key, but performance takes a backseat to cost. It's this arena that Priestley says was largely neglected under Intel's ownership.

To this end, much of Rivera's attention during the press briefing centered on emerging FPGA designs aimed at the mid-and embedded-market segments. This includes its Agilex 5 which brings with it a number of AI features embedded in the programmable logic.

This is a "very different approach from our competitors that have some AI blocks as part of their solution, but they use different tools, different flows. There's a new learning curve, and it's just harder to integrate," Rivera said in reference to AMD's Adaptive SoC designs which integrate general purpose Arm cores and AI accelerator blocks alongside a traditional FPGA.

That chip is slated to launch in April and, alongside Altera's Arria platform, is targeted at the middle of the FPGA market. Additionally, Rivera teased the Agilex 3, which is designed for low power, low cost, and small form factor applications typically seen in industrial environments. Rivera promised more details on those parts later this year.

A sticky situation

While building out these mid-to-low-end FPGA segments should make Altera's chips more attractive to a larger customer base, it may not be enough to convince long-time Xilinx customers to ditch AMD.

This is because, unlike in the CPU market, FPGAs are far more specialized and the applications built on them far less portable.

"When you pick one to go with, you're actually picking their design tools. So, your development team learns those design tools," Hoff explained, adding that this makes FPGAs rather sticky. Switching to another FPGA means learning that vendor's design tools, which can be a costly proposition, he added. As such Altera's growth is expected to be driven primarily by new and existing customers.

And when it comes to attracting new customers, Hoff suggests Altera double down on software development. "Cost of development and the time to market are really key drivers," he explained. The logic being that, if Altera's software is better or easier to work with, more folks will gravitate to their FPGAs.

Another opportunity Hoff sees is Intel's extensive intellectual property portfolio. "The more IP you can bring to your customers, the better, so they don't have to design everything."

• AMD crams five compute architectures onto a single board
• Microsoft adds FPGA-powered network accelerator to Azure
• AMD's latest FPGA promises super low latency AI for Flash Boy traders
• Intel spices up its FPGA game with open source and RISC-V freebies

Independence is a double-edged sword

Altera's position as a standalone FPGA biz represents something of a role reversal. Up until AMD's acquisition of Xilinx, that company was the sole pure-play FPGA vendor. Now, with the spinoff of PSG, Altera holds that crown.

"We are the only company left in the world that is a standalone FPGA company focused from top to bottom on FPGA capabilities," Rivera boasted to press.

But according to Priestley, this independence has its pros and cons. "It does give them the opportunity to to be more focused on the FPGA business and not subservient to the x86 market," he said.

On the flip side, Priestley emphasizes that the FPGA market is incredibly cyclic, which means big swings in demand. As Intel and AMD both noted in their most recent earnings calls, the FPGA market is currently undergoing an inventory correction which is depressing revenues.

While Altera was part of Intel, the company was insulated from these swings by the diverse nature of the chip giant's portfolio. As a standalone company, Rivera's Altera will be far more exposed.

The good news is there are a couple of developments on the horizon analysts predict could help Altera to gain a stable footing now that they've been cut loose.

According to Priestley, one of them is next-generation wireless tech. "In the past, we've seen sort of cyclical growth as the different set of technologies rolled out 3G, 4G, 5G that equated to FPGA growth," he said.

This makes the emerging 6G networking technologies a potential tailwind for Altera.

Another has to do with FPGAs used in the design of advanced semiconductors. With more fab facilities coming online and more companies opting to develop their own ASICs — many of which are apt to be fabbed in Intel foundries — it stands to reason Altera's FPGA sales should get a bump as well. Chip designs are often emulated on large banks of FPGAs to suss out any problems in the design prior to taping out.

However, at the end of the day, Priestley doesn't see the spinoff changing the fact that AMD's Xilinx remains the dominant FPGA vendor. ®

Now read: The Next Platform's analysis of the news and what it means for datacenters.