Intel has endured a strange few days after announcing its foundry business lost $7 billion last year, pushing its stock down by 12% in subsequent days. Some analysts predict Intel Foundry won’t dramatically improve until 2030.
Even so, the company is gushing about new semiconductors at Embedded World this week, including Core and Core Ultra processors for use at the edge as well as its Atom x7000C Series and Atom x7000RE Series. Also for the edge, Intel announced several customers using the new Arc GPU built-in to its Core Ultra: Jello (AI-enabled digital pathology), Sapient (digital signage), Iterate (GenAI chatbot for retail edge) and Shanghai Kaijing (object detection and classification).
Altera, a unit inside Intel, also announced Agilex 5 SoC FPGAs along with Quartus Prime v. 24.1.
Intel in a slide presentation said its “Agilex 5 fabric is infused with AI”—a reflection that AI is all around us these days. But does saying the fabric is ”infused’ only mean something metaphorical or something literal, as in a physical, technical process?
Intel answered that question of what is meant by “infused” in this context, implying it is a tech term. Here’s what an Intel spokesperson told Fierce Electronics:
“Our AI capability is built on the innovation of adding tensor modes within our Digital Signal Processing (DSP) Blocks. We distribute these blocks along with small blocks of memory, and the Adaptive Logic Modules (ALMs) that contain the Look Up Tables (LUTs) used as the basis of programmable logic, all connected with the Hyperflex enhanced programmable routing.
“This infusion of AI processing capability throughout the FPGA fabric removes bottlenecks between the ALMs, small memories, and the AI computations so that our FPGAi solution maximizes performance and minimizes resource usage. Software tools like FPGA AI Suite can use these blocks to create an AI processing IP that can be added to the rest of the processing pipeline. This combines the total solution of I/O capabilities like PCIe and memory, pre-and post-processing like video scaling, and AI processing like object detection to enable the workload agility our customers need.”
From that description, “infusion” is indeed meant as a technical term except when it isn’t—as when it’s marketing. (Big companies all do this.)
Whatever one thinks of the infusion definition, analyst Jack Gold of J. Gold Associates said the most important of Intel’s announcements is new FPGAs, which lately are the core components used in many acceleration jobs and not just AI. FPGAs have more performance and are used in a range of applications today, and the new Arc GPU will be important for embedded and edge devices for graphics applications and beyond.
Agilex 5 FPGA’s are now broadly available and Intel claims it has the performance per watt lead over competitors, and up to 2x better performance per watt than competing 7 nm FPGAs. That broad availability also means ecosystem partners can provide additional boards and system on modules.