As the major semiconductor companies tout ever more powerful GPUs and CPUs for AI training and inference, don’t sleep on the role of FPGAs in the AI revolution, as their flexible programmability and low latency features are helping them carve out a space in supporting a variety of AI workloads.
AMD highlighted its efforts in the FPGA sector this week with new hardware and software, and Intel throughout this year has engaged in a series of new FPGA product launches from its Programmable Solutions Group (PSG). That group, largely made up of assets from Intel’s 2015 Altera acquisition, was a source of much-needed positive news in Intel’s most recent earnings report, with a 35% year-over-year increase in revenue.
You can see why Intel this week at its FPGA Technology Day event launched even more new FPGA hardware and software as part of its growing Agilex family, with this latest round of products aimed at what the company believes is a very large, market-ripe portion of the FPGA market where many industrial use cases live.
“We're on the cusp now of our seventh consecutive quarter of record revenue year-on-year in PSG,” said Shannon Poulin, Corporate Vice President & General Manager of Intel’s PSG. “What you probably haven't seen from us are a lot of the roadmap changes that we've been making. We've invested in the very high end [FPGA] products, like for comms and cloud and very large infrastructure, since the acquisition of Altera... We've established a great position in that market. We haven't invested as much in the mid-range and low-end. Think about that as more industrial type applications and.. in the embedded space. We believe that they're great markets that we'll be able to penetrate.
The company is planning to test that belief with new devices for its line-ups of Agilex 3 and Agilex 5 FPGA devices. Here are the details:
-
Intel Agilex 3 FPGA Series: The company introduced B-Series and C-Series FPGAs to this group, which consists of power and cost-optimized FPGAs in compact form factors. They are essential building blocks targeted for a range of applications across markets, including system/board monitoring and management, video and vision, protocol expansion, portable imaging and displays, sensor fusion, drives, robotics I/O expansion and others. The B-Series FPGAs have higher I/O density in smaller form factors at lower power than Intel MAX 10 FPGAs, and target board and system management, including server platform management applications. The C-Series FPGAs offer added capabilities for a range of complex programmable logic devices and FPGA applications across vertical markets.
-
Intel Agilex 5 FPGA E-Series early access program expansion: Intel Agilex 5 FPGAs E-Series delivers cost-effective power and performance for embedded edge applications. E-Series FPGAs deliver up to 1.6 times better performance per watt compared to 16nm node competitors, Intel claimed. Power capabilities are accomplished using the second-generation Intel Hyperflex FPGA architecture combined with Intel 7 process technology, where transistors are optimized for performance per watt. Intel Agilex 5 FPGAs and SoCs take the industry’s first AI tensor block from Intel’s previous generation high-end offerings and bring them to mid-range FPGAs in Agilex 5 FPGAs, making them an ideal choice for edge AI applications. Intel is working with multiple customers designing E-Series devices as part of its early access program, with plans to begin sampling to early access customers in the fourth quarter of 2023. Intel will begin broad market shipments of E-Series engineering samples to customers in the first quarter of 2024, along with general access to design software. SIMICS, a complete system simulator for pre-silicon and post-silicon software development, testing and system integration, will be available for Agilex 5 with general access in the fourth quarter of 2023.
Among other FPGA news, Intel also said its Agilex 7 FPGAs with its R-Tile chiplet, which the company launched last spring, are now shipping.
In addition, Intel launched its Open FPGA Stack (OFS), an open source offering allowing developers full access to OFS hardware code, software code and technical documentation for platform and workload development. Open source OFS supports Intel Agilex FPGAs and Intel Stratix 10 FPGAs. Partners, including BittWare, Hitek Systems and SigmaX, have deployable OFS-based platform and application offerings available today, Intel said.