Welcome to the Altera Community!

A customer recently shared with me an interesting way they viewed the updated Altera brand: It’s like a long-time friend who had moved away for a few years but is now back and it’s time to get caught up. One of the things customers might want to 'catch up' on is Altera's efforts with regards to AI. It started when early FPGA products included the first basic digital signal processing (DSP) circuits within the FPGA fabric to improve performance for math-based logic, such as Fast Fourier transforms (FFTs) and finite impulse response (FIR) filters. These early enhancements improved general purpose FPGA-based computing but since 2015, our focus has shifted to improving AI capabilities in both silicon and software tools. DSP capabilities have gotten more sophisticated (fixed point, floating point, small and large bit precisions, etc.) and the quantity of available DSPs within a single device, have increased dramatically. Modern FPGAs are now capable of handling complex equations, especially those needed with the introduction of AI. This historical reminisce catches us up all the way until today’s news, where the latest Altera FPGA family is now broadly available to any customer who wants it; Agilex™ 5 SoC FPGAs, the first FPGAs infused with AI tensor blocks throughout the FPGA fabric. A short list of features that would be attractive to embedded or intelligent edge applications include: For those haven’t heard about Agilex™ 5 devices before today, here is how you can get started: Learn about the family. Review technical details. Download FPGA software. Free for anyone wanting access to Agilex™ 5 E-Series devices: Download Quartus® Prime Pro Test drive hardware (generally available now, lead-times may apply, via franchised distributors): Buy Altera development kits or 3rd party boards and SoMs The initial wave of board/SoM options include 9+ variants, with more coming. Evaluate AI or embedded options: Test out the FPGA AI Suite. Contact Altera sales for limited time introductory pricing. Utilize 3rd party tools (Arm DS, MathWorks) to design for the new, best-in-class Arm dual-A76 + dual-A55 based SoC subsystem or RISC-V based Nios® V soft IP processors. Altera is announcing the Agilex™ 5 family broad availability coincident with Embedded World 2024 because it is one of the key markets this mid-range FPGA family was architected for. Embedded customers clearly told us they perceive a lack of adequate compute in embedded processors, see a big need to fill security gaps, and want to add AI into their next generation systems. Agilex™ 5 devices can address all these concerns. Coming back to our initial topic, AI: GPUs are certainly a popular choice for AI training, but power consumption of GPUs for AI inferencing may be too high for intelligent edge or embedded applications. Instead of adding a separate GPU/AI semiconductor device to an embedded system (resulting in higher cost, more power, more thermal, etc.), why not add the AI function into an FPGA already planned to be used in embedded/edge equipment? For decades, FPGAs have been used in embedded/edge and communication systems for real-time control, IO connectivity, or image/data processing. The estimates on Agilex™ 5 device AI performance look good compared to equivalent class competitors. Because the FPGAs new DSP/tensor is implemented in a fine-grained architecture, it provides the FPGA designer the ability to tune for higher performance or lower power consumption, using the minimum amount of FPGA resources for the desired algorithm. Agilex™ 5 devices – AI key figures of merit: Tensor neural acceleration performance: Up to 26 / 56 TOPS ² Better performance per power efficiency versus embedded market inference GPUs. 1.7x higher frames per second per watt ³ Better raw performance versus other AI targeted FPGAs. 69% higher frames per second ⁴ There are many great reasons to look at this new family of FPGAs. If you are an architect, AI developer, or FPGA designer for embedded systems, don’t wait. As Mark Twain famously said, “The secret of getting ahead is getting started.” Don’t believe the marketing hype, try out your ideas in actual hardware, to see what is ‘possible’. Agilex™ 5 SoC FPGAs are just the latest phase in our DSP/AI journey. Altera, accelerating innovators. Come visit us at Embedded World 2024: Altera booth Hall 5, 5-135 and 5-136. Footnotes: Performance per watt: https://edc.intel.com/content/www/us/en/products/performance/benchmarks/agilex-fpga/ Theoretical peak INT8 calculations for the largest density Agilex 5 E-Series or D-Series devices. 1.7x higher frames per second per watt vs. Nvidia Jetson-class GPUs (AGX Orin) 69% higher frames per second vs. AMD/Xilinx Versal AI devices (VE2302)

As industries race to unlock real-time insights from massive volumes of sensor data, the need for high-performance, low-latency computing at the edge has never been greater. From medical imaging to industrial automation and autonomous robots, success depends on the seamless integration of data capture, processing, and AI-driven decision-making. That’s why Altera is proud to be an ecosystem partner of NVIDIA Holoscan, working together to enable a new class of accelerated computing platforms designed for sensor-rich, AI-powered applications. Leading the Way in High-Speed Connectivity Innovation at the edge starts with moving data fast and reliably. Altera is proud to offer the FPGA industry’s only 25G and 100G Holoscan Sensor Bridge designs, providing the high-bandwidth infrastructure needed to handle today’s most demanding Edge AI workloads. These high-speed interfaces are critical for: Streaming multiple high-resolution video or other high-bandwidth data streams Processing large radar, robotics, drone, or medical imaging datasets Enabling low-latency, deterministic decision-making in mission-critical environments “With Altera FPGAs driving these capabilities, developers can scale their applications confidently, knowing their systems won’t be bottlenecked by sensors and data movement pipelined via Holoscan Sensor Bridge into NVIDIA GPUs,” says Farhad Shafai, Altera, Head of Business Solutions and Vertical Marketing. Why Altera FPGAs + GPUs Are Better Together Modern intelligent systems require both flexibility and performance, which is why combining Altera FPGAs and GPUs is becoming the architecture of choice. Here’s how this powerful pairing works: FPGA Capability How It Enhances GPU-Based AI Sensor Connectivity and Data Ingestion FPGAs excel at interfacing directly with diverse sensors, handling custom protocols, and aggregating data streams efficiently. Ultra-Fast Processing With deterministic, low-latency performance, FPGAs can preprocess and filter data before it reaches the GPU, ensuring only the most relevant information is passed along. AI Acceleration FPGAs can be designed into your solution to pre-process and prepare your data for the most efficient AI processing by the GPU. Flexibility and Futureproofing FPGAs offer reconfigurability, allowing systems to adapt to evolving standards and algorithms without redesigning hardware. Enhanced Security With built-in hardware-level security features, FPGAs can encrypt sensor data directly at the source, helping ensure that sensitive information is protected from the edge to the compute/GPU, critical in industries like defense, and industrial automation. Together, Altera FPGAs and NVIDIA accelerated computing create an architecture that delivers both performance and adaptability, enabling developers to build smarter, faster, and more efficient systems. See It in Action: Altera’s 25G Demo on Agilex® 5 SoC FPGAs Want to see how high-speed connectivity and intelligent processing come together in real-world applications? Watch the demo: https://www.youtube.com/watch?v=ikmNKdOHbUU This demonstration highlights Altera’s latest 25G Holoscan Sensor Bridge design, enabling multi-camera sensor data flow and accelerated processing at the edge. Ready to Get Started? Developers can jump in quickly and start building with our 10G, 25G, and 100G reference designs at: Get started on GitHub https://github.com/altera-fpga/holoscan-sensor-bridge/tree/altera-release-2.6.0/fpga/altera Whether you’re building next-generation medical devices, robotics platforms, or AI-powered industrial systems, Altera and NVIDIA Holoscan provide the foundation you need to succeed. Learn More Explore Altera’s 10G – 100G Holoscan Sensor Bridge offerings at:  http://www.altera.com/holoscan-sb

Smart cameras and embedded vision systems are being asked to do more at the edge. They need to capture higher-resolution video, process images in real time, support evolving sensors and interfaces, and prepare clean data for analytics or AI. At the same time, many of these systems are constrained by board area, system cost, power budgets, and long product lifecycles. That combination creates a familiar engineering challenge: how much of a real vision pipeline can be built in a device class optimized for lower logic density and cost? The Agilex® 3 4Kp30 Camera Lite reference design gives a compelling answer. It demonstrates a practical sensor-to-display camera pipeline built on Agilex 3, showing how a power- and cost-optimized FPGA can ingest a 4K image stream over MIPI CSI-2, process raw sensor data through an image signal processing pipeline, and output 4Kp30 video through DisplayPort 1.4. The reference design is a complete working vision pipeline that you can use to build your own based on your unique requirements. From sensor input to display output The reference design starts with a Raspberry Pi High Quality Camera module using the Sony IMX477 image sensor. The sensor outputs 12-bit raw Bayer data and connects to the FPGA through a MIPI CSI-2 interface. From there, the MIPI CSI-2 IP converts the incoming pixel stream into AXI4-Streaming data, making it available to the rest of the Altera® Video and Vision Processing (VVP) Suite pipeline. Inside the FPGA fabric, the design implements the major stages needed to turn raw sensor data into display-ready video. The ISP pipeline includes Black Level Correction, White Balance Correction, Demosaic, Color Correction Matrix, and a 1D LUT. The pipeline supports 12-bit raw data up to the Demosaic IP and 10-bit RGB for downstream video processing. The result is a fixed 3840 x 2160, 30 Hz video path from camera input to display output. That matters because many real products need more than a way to receive camera data. They need image correction, color processing, buffering, video formatting, display output, and software control. Agilex 3 brings those pieces together in a reference design that engineers can study, run, and adapt. A practical foundation for smart camera products For security cameras, industrial vision, smart infrastructure, robotics, retail analytics, and access control, the camera pipeline is often the first major design decision. The system must bring pixels in from the sensor, correct and format them, keep timing deterministic, and deliver data to the next stage of the product. In many cases, that next stage may be a display path, a host processor, a networking subsystem, a storage path, or an AI analytics engine. The Agilex 3 reference design is valuable because it gives developers a working foundation for that pipeline. It demonstrates the sensor input, image-processing path, video frame buffer, output mixer, and DisplayPort output using Altera IP. Developers can use it as a starting point, then adapt the pipeline for their own sensor choice, image-processing requirements, overlay needs, output path, or product-specific differentiation. In the current reference design, the video frame buffer is used for video synchronization. This is a practical detail worth highlighting because real video designs often need buffering for timing alignment, rate matching, format conversion, or system-level processing. The design uses external LPDDR SDRAM through an external memory interface for that frame buffer. More broadly, Agilex 3 SoC devices also supports LPDDR5 memory interfaces, giving production designs a path to compact memory subsystems around video and embedded processing workloads. Embedded control today, hard processor path for production systems The current Agilex 3 Camera Lite reference design uses a Nios® V soft processor running a bare-metal software application. That software discovers the hardware IP blocks, configures them, monitors feedback loops, and provides a terminal-based interface over JTAG-UART. This is a good fit for the reference design because it keeps the example compact and focused on the FPGA-resident video pipeline. For production smart camera or industrial vision systems, Agilex 3 SoC devices add another important platform option: an integrated hard processor system with dual-core Arm® Cortex®-A55 processors. That HPS is not used in this reference design, but it can be a major advantage in a product architecture. Designers can use the FPGA fabric for deterministic video ingest and image processing while using the HPS for system control, application software, communication stacks, user interfaces, sensor orchestration, security services, or higher-level product logic. This combination is especially important for embedded vision. Hardware pipelines are excellent at moving and processing pixels predictably. Software is excellent at managing the product around that pipeline. Agilex 3 gives designers both paths in the same device family: a fabric-based video processing foundation and, when needed, an integrated Arm-based processing subsystem for production software. A path toward AI-enhanced edge vision The Agilex 3 4Kp30 Camera Lite reference design is not an AI inference design. Its focus is 4K camera ingest, ISP processing, video buffering, output mixing, and DisplayPort output. That distinction is important. At the same time, the architecture points naturally toward smarter edge vision systems. Agilex 3 includes AI-capable Tensor Block architecture in the fabric, and Agilex 3 and Agilex 5 share an architecturally aligned FPGA fabric foundation. The related Agilex 5 camera reference design shows the fuller concept by combining multi-sensor 4K camera input, ISP processing, FPGA AI Suite inference, Linux software on the HPS, and display output with AI results. That gives customers a scalable design story. Agilex 3 can be the cost-optimized starting point for 4K smart camera pipelines and edge vision preprocessing. Agilex 5 can scale the concept to larger, more compute-intensive designs that integrate AI inference directly into the reference architecture. Customers can begin with the 4K vision pipeline they need today and scale toward more intelligent camera systems as product requirements evolve. Why it matters The real message of the Agilex 3 4Kp30 Camera Lite reference design is: a lower-density, cost-optimized FPGA can still implement a substantial portion of a modern vision system using available Altera IP. That includes native camera ingest, AXI4-Streaming video movement, ISP processing, frame buffering, output mixing, embedded software control, and DisplayPort output. For customers building smart cameras, industrial vision systems, surveillance endpoints, or edge AI preprocessing pipelines, this is a practical starting point rather than a blank page. With Agilex 3, designers can build compact, customizable 4K vision systems while keeping a clear path to hard processor integration, modern memory support, and future AI-enhanced processing. It is a strong example of how much capability can fit into the power- and cost-optimized segment of the Agilex portfolio. Explore the Agilex 3 4Kp30 Camera Lite reference design and use it as a starting point for your next smart camera, industrial vision, surveillance, or edge AI preprocessing system. Source links for reviewers Agilex 3 4Kp30 Camera Lite developer documentation Agilex 3 camera GitHub repository Agilex 3 FPGAs and SoCs device overview Agilex 3 HPS documentation Agilex 5 4Kp30 Multi-Sensor Camera with AI Inference documentation

Engineers starting a new FPGA design make several important choices before they write the first line of RTL. The device they select influences the tool flow, IP base, debug method, training path, and production roadmap that follow. For teams working on cost-sensitive embedded, vision, control, DSP, and edge AI-enabled systems, the ideal starting point needs to be affordable, accessible, and serious enough to support real implementation work. Agilex® 3 brings that starting point into the modern Agilex portfolio. Built with Intel 7 technology and second-generation HyperFlex™ architecture, Agilex 3 FPGAs and SoCs extend Agilex-class fabric, efficient performance, modern I/O, embedded memory, DSP resources, and security features into power- and cost-optimized applications. The result is a practical entry point for developers who want a current-generation FPGA platform with a path that can grow beyond the first evaluation board. Start with a real Agilex device for less than $130 The Arrow AXC3000 starter kit gives developers a low-cost way to begin designing with Agilex 3. Currently listed through Arrow at $129, the kit is based on the Agilex 3 C-Series 100 device A3CY100BM16AE7S. That means developers can start with a real Agilex 3 FPGA that provides roughly 100K logic elements, 138 DSP blocks, 276 18x19 multipliers, MIPI D-PHY support, LVDS, a Secure Device Manager, internal memory, external HyperRAM, QSPI configuration flash, CRUVI HS expansion, Arduino MKR standard pads, USB-C power, and an on-board programmer and debugger. This gives first-time Agilex 3 users a compact board for labs, proofs of concept, and early design exploration while preserving a connection to production-capable architecture and tools. It also gives experienced teams a low-cost evaluation path for assessing Agilex 3 in edge, industrial, vision, and control applications. The hardware ecosystem gives developers several ways to start. In addition to AXC3000, Terasic lists Agilex 3 kits under $180, including the DE23-Lite Development Kit, Atum A3 Nano, and Atum Nios® V Starter Kit. These options support different entry paths: education and prototyping, compact FPGA development, and Nios V-focused embedded exploration. A complete set of professional tools for no cost The low-cost Agilex® 3 makes FPGA development more accessible, while a no-cost professional tool flow helps developers realize the device’s full performance and design potential. Quartus® Prime Pro Edition software is the professional design environment and is available from the Altera Download Center. For Agilex 3 devices, developers can use Quartus Prime Pro with a no-cost license, providing a fast path from installation to first compile without a separate manual licensing workflow. Quartus Prime Pro gives Agilex 3 developers a professional FPGA implementation environment from day one. The integrated flow includes synthesis, place-and-route, timing analysis, system debug, power analysis, IP integration, and example designs. Altera knows that FPGA designs can become complex quickly, even during early evaluation. Developers need strong visibility into design behavior so they can find issues, validate functionality, and keep projects moving. For that reason, Altera does not charge for system debug tool. We provide a full license for no cost for any Agilex 3 design. Developers can also use advanced Altera tools and capabilities such as DSP Builder, Power & Thermal Analyzer, FPGA AI Suite, and Nios V processor IP where they fit the design objective. That matters because the first evaluation project often becomes the foundation for production work. Design teams want to learn a flow that can carry forward. They want timing closure methods, IP, debug tools, and software habits that remain useful as the design grows. Agilex 3 gives them that continuity in a professional Quartus Prime Pro environment. Application-ready examples make the story concrete The most compelling way to evaluate a starter platform is to see it run a meaningful application. Agilex 3 is well suited for that kind of demonstration because it combines modern FPGA fabric, useful I/O, embedded memory, DSP resources, and application-oriented IP in a cost-optimized device family. A strong example is the Agilex 3 4Kp30 camera reference design. The design shows how a compact FPGA platform can support a complete video pipeline, including image sensor input, MIPI-based connectivity, image signal processing, and display output. For vision developers, that is a much more useful starting point than a generic board bring-up exercise. It connects directly to applications such as industrial vision, smart cameras, robotics, surveillance, retail analytics, and edge preprocessing. For the AXC3000 mentioned above, a GitHub repo is provided that includes several designs for the AXC3000 board: https://github.com/ArrowElectronics/Agilex-3/wiki/Agilex-3-AXC3000-Development-Platform#reference-designs The same principle applies to embedded and software-oriented development. Nios V provides a RISC-V processor path for control-plane and embedded workflows, and the Atum Nios V Starter Kit gives developers a focused way to explore that environment. As additional Agilex 3 software examples become ready for public promotion, they can extend the story into platform management, real-time control, and HPS-based use cases. A starter path that scales Agilex 3 is the accessible on-ramp to the broader Agilex portfolio and Especially Agilex 5 E-series. Developers can begin with low-cost starter hardware, use professional Quartus Prime Pro tools, and build confidence on real application examples. As designs become more demanding, the path does not stop at the entry board. More capable Agilex 3 boards and device options give teams room to expand into richer I/O, larger designs, more embedded memory, and more complete application prototypes. For software-rich embedded systems, Agilex 3 SoC devices extend the story further with a dual-core Arm Cortex-A55 hard processor system integrated alongside the FPGA fabric. That gives developers a path from simple FPGA evaluation into more powerful applications that combine programmable logic, embedded software, real-time control, and system-level processing in one platform. Need more gates? Even more powerful devices? A pin to pin migration path to Agilex 5 E-series will get you there. The same design environment can support early prototyping, design optimization, IP integration, and eventual deployment across Agilex 3 devices and the broader Agilex family.

How a published OpenCores study connects workload behavior, fabric architecture, and software optimization. Mid-range FPGA designers rarely optimize for only one thing. A design has to fit. It has to close timing. It has to stay within the power budget. As more functions move into the same device, those requirements become harder to satisfy at the same time. That is why the combination of fabric architecture and software matters. A new Altera white paper, ”Altera Delivers Superior Performance, Power, and Logic Packing with Quartus® and Agilex® 5 FPGAs Versus AMD Kintex™ UltraScale+”, examines this challenge using twelve publicly available OpenCores benchmark designs. The study compares Agilex® 5 E-Series devices with AMD Kintex™ UltraScale+ devices using each vendor’s recommended software flow and a stamping methodology that increases utilization by replicating each design across the FPGA. FPGA performance is application-dependent, and the paper’s value is in showing how different workload types stress different parts of the FPGA architecture and implementation software, highlighting where Agilex 5 E-Series demonstrates meaningful advantages. Why Agilex 5 E-Series is well suited for these workloads Agilex 5 E-Series is designed for mid-range applications that need a balance of performance, power efficiency, and integration density. Its fabric combines HyperFlex™ architecture, adaptive logic modules, embedded M20K memory, and Quartus® Prime optimization technology. Just as important, its regular fabric structure and scalable interconnect help routing remain more predictable as utilization rises. That matters because timing closure is often limited by placement and routing pressure, not only by raw logic count. Because Agilex 5D shares the same underlying Agilex 5 fabric architecture, these fabric-level advantages extend to the D-Series as well, with higher device density, greater on-chip resource capacity, and approximately 20% higher fabric performance. Where the advantages show up For designs where routing and timing closure dominate, such as deeply pipelined processor and error-correction workloads, Agilex 5 E-Series benefits from HyperFlex retiming and register duplication. These capabilities help break up long interconnect paths, reduce critical path depth, and sustain fMAX as replicated design instances push utilization higher. For arithmetic-heavy workloads, such as trigonometric, DSP, and iterative compute kernels, Agilex 5 adaptive logic modules help pack arithmetic more locally and efficiently. Better local packing can reduce routing overhead, which becomes increasingly important as dense compute designs scale across the device. For memory-sensitive and resource-constrained workloads, including video, security, and embedded compute designs, embedded memory and logic packing become key limiters. Agilex 5 E-SeriesM20K resources and efficient ALM utilization help preserve usable logic capacity and timing margin when competing implementations may need more LUT-based memory or additional routing resources. For designs where implementation efficiency depends heavily on the tool flow, Quartus Prime advanced synthesis and physical synthesis add another part of the formula. The software can co-optimize logic, placement, and routing, helping improve utilization and power efficiency without requiring RTL changes. The proof, and what it means The published white paper provides the detailed benchmark data, but several results stand out because they connect directly to real design challenges: 22% higher geometric mean performance at high-utilization operating points. This matters because FPGA designs often become harder to close as utilization rises. Higher performance under those conditions points to better timing behavior when routing pressure, placement density, and resource contention increase. More stable achievable fMAX as utilization increases. The paper shows less frequency degradation for Agilex 5 E-Series as benchmark instances are stamped across the device. For designers, that means more predictable timing closure as the design approaches fuller device usage. 47% lower total power at iso-frequency and iso-workload conditions. This result isolates power at equivalent workload and clock frequency, showing that Agilex 5 E-Series can do the same work with significantly lower total device power in the evaluated designs. Up to 2.83x greater energy efficiency in individual workloads. This shows that the advantage is not only about reducing watts. It also reflects better work-per-watt behavior, which can help improve thermal margin, power budgets, and system-level efficiency. 17% higher effective device capacity. In this context, effective capacity reflects the amount of design logic that can be implemented while still meeting timing constraints Read the full white paper This blog highlights the key findings, but the full whitepaper provides detailed benchmark data, workload-by-workload analysis, methodology transparency, and architectural insights across all twelve OpenCores designs. Read the whitepaper to understand how Agilex 5 E devices and Quartus Prime software delivered higher performance, lower power, and greater effective design capacity across a diverse set of real FPGA workloads. Altera Delivers Superior Performance, Power, and Logic Packing with Quartus® and Agilex® 5 FPGAs Versus AMD Kintex™ Ultrascale+