News

Intel Spins Off Altera as Independent FPGA Supplier

March 08, 2024 by Duane Benson

Eight and a half years after acquiring Altera, Intel has spun off the FPGA division as an independent subsidiary.

Intel recently announced the launch of Altera as a standalone FPGA company. Intel purchased Altera in 2015 to diversify its portfolio, enter the programmable logic world, and expand its data center computing offerings. As an Intel division, it was named the Programmable Solutions Group.

 

Altera’s FPGAs

Altera’s FPGAs integrate AI inferencing capabilities and intercept standards like PCI Express, CXL, Ethernet, and 6G.
 

While the new Altera company still has a portfolio of advanced field programmable gate arrays (FPGAs), it also has products and features that target the complex needs of today’s market, including multiple variations of AI, edge computing, and high-bandwidth, mixed-signal electronics.

 

Newest Altera Products

Altera's latest Agilex products cover a broad spectrum of applications—from low-power, low-cost embedded edge computing to high-end, mixed-signal data and processing FPGAs. 

 

Agilex line optimizations

Agilex line optimizations. 

 

Agilex 9 FPGAs are designed for RF applications and include wideband data converters with sample rates up to 64 Gsps. Agilex 7 (F series and I series) is optimized for high-bandwidth applications like data centers, defense, and networking. It delivers 2x higher performance per watt. Agilex 5 comes with AI capability built-in and a promised 1.6x better-per-watt performance than competitive products. The Agilex 3 line is on deck for release next and is targeted at low-power applications that have less complex needs, like cloud, communications, and edge computing.

 

Agilex 5 and the AI Surge

Graphics processing units (GPUs) and tensor processing units (TPUs) may be the heavy engines of AI large language model (LLM) processing, but that doesn’t mean there’s no room for FPGAs in AI. 

FPGAs can impact applications where significant amounts of data must be pre-processed, including analog conversion. There’s also an FPGA opportunity in edge AI and areas where the “final” hardware is not as final as the developers would like. FPGAs allow system designers to ship products with the ability to upgrade their hardware in the future.

The Agilex 5 series targets AI and edge-centric applications that need more processing power and greater diversity of data handling than what a conventional CPU, GPU, or TPU can offer. The Agilex 5 series FPGAs are designed on Intel process 7 (10 nm Enhanced SuperFin), which emphasizes performance per watt. In addition, the 5 series contains DSP and tensor processing blocks built in. These blocks are improved over previous versions, giving better performance per watt with the same die size.

 

Block diagram of Agilex 5 FPGAs and SoCs

Block diagram of Agilex 5 FPGAs and SoCs.
 

Intel has also developed chiplet FPGA fabric, giving designers the ability to add FPGA capability to SoCs and other processing chips. The parallel nature and flexibility of an FPGA make it an useful add-in for advanced data management.

 

Expanding Application Potential With Agilex 3

The Agilex 3 family has a smaller form factor, lower costs, and lower power consumption than the Agilex 5. As such, it is designed for applications even closer to the edge, such as mobile devices, robotics, and power-critical components in data center and server applications. The Agilex 3 succeeds the Intel MAX 10 FPGA family, while the Agilex 5 series is the successor to the Cyclone V, Intel Cyclone 10, and Intel Arria 10 families.

 

Development Tools

Altera offers a set of development kits and an improved development toolchain. The kits and toolchains are designed to help engineers start on their designs sooner, ultimately reducing the product development time cycle. 

In addition to the standard FPGA Quartus design suite, Intel offers the Intel FPGA AI suite for AI-targeted applications. The AI suite uses AI blocks similar to a soft-core processor block or any other IP set to make adding AI processing capability to an FPGA’s fabric familiar to most seasoned FPGA developers.

 

What’s Next for Altera?

When Intel acquired Altera in 2015, its intended focus was on data centers and the Internet of Things (IoT). For the new Altera, high-end FPGA applications continue to be an important market segment. With AI built into Agilex 5 and the lower cost and power footprints of Agilex 3, Altera has an answer to the explosion of AI computing, both in the cloud and on the edge.