NVIDIA’s desktop team presents DGX Spark as a true “personal AI supercomputer” form factor: a Grace Blackwell GB10 Superchip system delivering around 1 petaFLOP of FP4 AI performance, 128 GB of coherent unified memory and 4 TB NVMe storage so LLMs and multimodal models up to roughly 200 billion parameters can run locally instead of in the cloud. With the full NVIDIA AI software stack preinstalled and a ConnectX-7 SmartNIC for 100 GbE plus 10 GbE RJ-45, two DGX Spark units can be linked as a mini cluster to push into 400B-parameter experimentation while staying on the developer’s own desktop. https://www.nvidia.com/en-us/products/workstations/dgx-spark/

—
HDMI® Technology is the foundation for the worldwide ecosystem of HDMI-connected devices; integrated with displays, set-top boxes, laptops, audio video receivers and other product types. Because of this global usage, manufacturers, resellers, integrators and consumers must be assured that their HDMI® products work seamlessly together and deliver the best possible performance by sourcing products from licensed HDMI Adopters or authorized resellers. For HDMI Cables, consumers can look for the official HDMI® Cable Certification Labels on packaging. Innovation continues with the latest HDMI 2.2 Specification that supports higher 96Gbps bandwidth and next-gen HDMI Fixed Rate Link technology to provide optimal audio and video for a wide range of device applications. Higher resolutions and refresh rates are supported, including up to 12K@120 and 16K@60. Additionally, more high-quality options are supported, including uncompressed full chroma formats such as 8K@60/4:4:4 and 4K@240/4:4:4 at 10-bit and 12-bit color.
—

In the interview, NVIDIA positions DGX Spark as a way for researchers and engineers whose laptops are hitting memory and bandwidth limits to keep working inside familiar IDEs and toolchains while offloading training, fine-tuning, RAG and agent workloads to a silent 1.2 kg box beside the monitor. The industrial design deliberately echoes the original DGX-1 Jensen Huang delivered to OpenAI, but compresses a rack-class capability into a footprint of 150 x 150 x 50.5 mm, turning what used to require a lab into something that fits on a shared office desk.

Next to DGX Spark, NVIDIA showcases its RTX Pro 6000 Blackwell workstation GPUs, in Max-Q 300 W, 600 W active-cooled desktop and passive server variants. Built on the Blackwell architecture with 96 GB of GDDR7, a 512-bit memory interface and roughly 1.8 TB/s bandwidth, these boards bring fifth-generation Tensor Cores with FP4/FP8 modes and fourth-generation RT Cores into single-socket workstations, with support for four-GPU configurations over PCIe Gen 5. For CAD, DCC, digital twins, robotics simulation and small-to-mid-scale model training, that combination of VRAM capacity and throughput is what keeps data on-GPU instead of constantly paging to system memory or the network.

Recorded at SC25 in St Louis, the conversation ties these pieces together as a continuum from desktop AI boxes to multi-GPU workstations and datacenter nodes, all sharing the same Blackwell, CUDA and NVIDIA AI Enterprise software ecosystem. OEM partners like Dell, HP, Lenovo, MSI and others are already integrating RTX Pro 6000 Blackwell and Grace Blackwell platforms into their own systems, so developers can choose between a turnkey DGX Spark on the desk, a full workstation packed with RTX Pro 6000 boards under the desk, or scaling out to clusters while keeping a consistent toolchain across the whole AI community.

I’m publishing about 90+ videos from Embedded World North America 2025, I upload about 4 videos per day at 5AM/11AM/5PM/11PM CET/EST. Join https://www.youtube.com/charbax/join for Early Access to all 90 videos (once they’re all queued in next few days) Check out all my Embedded World North America videos in my Embedded World playlist here: https://www.youtube.com/playlist?list=PL7xXqJFxvYvjgUpdNMBkGzEWU6YVxR8Ga

This video was filmed using the DJI Pocket 3 ($669 at https://amzn.to/4aMpKIC using the dual wireless DJI Mic 2 microphones with the DJI lapel microphone https://amzn.to/3XIj3l8 ), watch all my DJI Pocket 3 videos here https://www.youtube.com/playlist?list=PL7xXqJFxvYvhDlWIAxm_pR9dp7ArSkhKK

Click the “Super Thanks” button below the video to send a highlighted comment under the video! Brands I film are welcome to support my work in this way 😁

Check out my video with Daylight Computer about their revolutionary Sunlight Readable Transflective LCD Display for Healthy Learning: https://www.youtube.com/watch?v=U98RuxkFDYY

source https://www.youtube.com/watch?v=3xu4XuH6MFU

SALZ Automation shows how virtualization and functional safety can run on the same industrial PC. With its SALZ Control System and ZOS real-time Linux, the company runs CODESYS Virtual Safe Control SL as a SIL3 safety controller on standard Advantech edge hardware instead of dedicated safety PLCs, turning commercial off-the-shelf systems into certified safety platforms. https://www.salz-automation.com/en/From_Virtualisation_to_Functional_Safety

The demo implements an emergency-stop scenario using Functional Safety over EtherCAT (FSoE) and mixed I/O vendors, with virtual PLC and virtual safe PLC instances replacing classic hardwired safety CPUs. Safety logic is engineered in CODESYS but executed inside containers, where safe runtime and standard control runtime are isolated and monitored via the SALZ dashboard while the full setup draws only about six percent CPU on the Advantech box.

Architecturally, SALZ maps the CODESYS “Safe House” pattern into a containerized system: one host provides an external time source, the other runs the virtual safe controller and standard control tasks as dual software channels that meet IEC 61508 SIL3 timing requirements. This virtualized safety path preserves deterministic execution and diagnostic coverage while replacing hardware redundancy with software-defined redundancy managed through Docker-style orchestration.

Because the stack is protocol-agnostic, the same approach can be applied in robotics cells, packaging lines, food and beverage plants or process and gas applications wherever functional safety and flexible topology are required. OEMs and integrators can scale from single machines to fleets by spinning up containers instead of redesigning hardware, while staying in the standard CODESYS Safety engineering environment.

Recorded at the Advantech booth during SPS 2025 in Nuremberg, the discussion positions this as an early live deployment of virtual safety control on generic edge hardware. It reflects a broader shift in industrial automation from fixed, hardware-bound safety PLCs to software-defined control on open PCs, with containerization, EtherCAT and OPC UA helping virtual PLCs and virtual safe controllers coexist on the same box without compromising certified safety levels.

source https://www.youtube.com/watch?v=pJ8JsbagAHk

Axiomtek showcases how its own motherboard-to-system design strategy scales from smart city video walls to edge AI inspection in this tour of their Intel and NVIDIA powered platforms. The demo starts with a smart city monitoring system that can drive up to nine simultaneous video streams, using dedicated acceleration to handle high-bandwidth analytics rather than simple passive display. https://www.axiomtek.com/

—
HDMI® Technology is the foundation for the worldwide ecosystem of HDMI-connected devices; integrated with displays, set-top boxes, laptops, audio video receivers and other product types. Because of this global usage, manufacturers, resellers, integrators and consumers must be assured that their HDMI® products work seamlessly together and deliver the best possible performance by sourcing products from licensed HDMI Adopters or authorized resellers. For HDMI Cables, consumers can look for the official HDMI® Cable Certification Labels on packaging. Innovation continues with the latest HDMI 2.2 Specification that supports higher 96Gbps bandwidth and next-gen HDMI Fixed Rate Link technology to provide optimal audio and video for a wide range of device applications. Higher resolutions and refresh rates are supported, including up to 12K@120 and 16K@60. Additionally, more high-quality options are supported, including uncompressed full chroma formats such as 8K@60/4:4:4 and 4K@240/4:4:4 at 10-bit and 12-bit color.
—

On the NVIDIA side, Axiomtek’s AI series integrates Jetson-based compute into rugged fanless boxes with GMSL-style camera connectors, so integrators can bring in four or eight cameras per node depending on the configuration. The suitcase-counting demo illustrates how multi-camera computer vision, object detection and tracking can be deployed directly at the edge for logistics, retail or transportation analytics without sending all the raw video to the cloud.

A second demo focuses on anomaly detection and quality inspection, where Axiomtek’s AIE500 hardware runs a partner’s AI software stack. Operators capture reference images of a “good” bolt or PCB, then the model learns to flag deviations on the production line or in agriculture and other verticals. The system exposes confidence scores and thresholds, so process engineers can tune false positive and false negative rates for their specific workflow rather than treating AI as a black box.

Behind the glass wall, Axiomtek walks through its “bread and butter” embedded boards, from full-size ATX all the way down to the credit-card-sized KIWI300 platform. These boards target transportation, networking, retail, gaming and industrial control, and feed into DIN-rail IoT gateways such as the IECO and IIA series. In this Embedded World North America 2025 setup, the gateways act as secure edge nodes, handling connectivity, cybersecurity policies and protocol translation between field devices and higher-level IT systems.

The tour ends on industrial panel PCs and Intel-powered embedded systems designed as HMIs and edge analytics nodes for factories and smart buildings. A proximity and ambient-light sensor automatically dims the display and reduces power draw when nobody is nearby, then wakes the interface when an operator approaches. A security-oriented demo shows how vision analytics can detect people lingering in a zone, similar to a professional “ring camera” that can trigger alerts or escalation. With manufacturing in Taiwan and integration and engineering support from its City of Industry office in the US, Axiomtek positions these platforms as building blocks for scalable, production-ready edge AI deployments.

I’m publishing about 90+ videos from Embedded World North America 2025, I upload about 4 videos per day at 5AM/11AM/5PM/11PM CET/EST. Join https://www.youtube.com/charbax/join for Early Access to all 90 videos (once they’re all queued in next few days) Check out all my Embedded World North America videos in my Embedded World playlist here: https://www.youtube.com/playlist?list=PL7xXqJFxvYvjgUpdNMBkGzEWU6YVxR8Ga

This video was filmed using the DJI Pocket 3 ($669 at https://amzn.to/4aMpKIC using the dual wireless DJI Mic 2 microphones with the DJI lapel microphone https://amzn.to/3XIj3l8 ), watch all my DJI Pocket 3 videos here https://www.youtube.com/playlist?list=PL7xXqJFxvYvhDlWIAxm_pR9dp7ArSkhKK

Click the “Super Thanks” button below the video to send a highlighted comment under the video! Brands I film are welcome to support my work in this way 😁

Check out my video with Daylight Computer about their revolutionary Sunlight Readable Transflective LCD Display for Healthy Learning: https://www.youtube.com/watch?v=U98RuxkFDYY

source https://www.youtube.com/watch?v=O4J61o1QASM

Hybrid quantum-classical computing is moving from theory into real workflows, and IBM is using this demo to show how that looks on IBM Cloud. A single application is split into several stages: classical HPC jobs prepare and analyze molecular configurations, then a quantum circuit is generated and executed on an IBM quantum processor, before the final results are visualized for the user. IBM Spectrum LSF acts as the batch scheduler and control plane that orchestrates these stages end to end on cloud infrastructure. https://www.ibm.com/products/hpc-workload-management

—
HDMI® Technology is the foundation for the worldwide ecosystem of HDMI-connected devices; integrated with displays, set-top boxes, laptops, audio video receivers and other product types. Because of this global usage, manufacturers, resellers, integrators and consumers must be assured that their HDMI® products work seamlessly together and deliver the best possible performance by sourcing products from licensed HDMI Adopters or authorized resellers. For HDMI Cables, consumers can look for the official HDMI® Cable Certification Labels on packaging. Innovation continues with the latest HDMI 2.2 Specification that supports higher 96Gbps bandwidth and next-gen HDMI Fixed Rate Link technology to provide optimal audio and video for a wide range of device applications. Higher resolutions and refresh rates are supported, including up to 12K@120 and 16K@60. Additionally, more high-quality options are supported, including uncompressed full chroma formats such as 8K@60/4:4:4 and 4K@240/4:4:4 at 10-bit and 12-bit color.
—

The video walks through a typical hybrid workload for quantum chemistry and materials modeling, where classical nodes first determine the geometry of a molecule and perform the heavy pre-processing. A subsequent step transposes and lays out the quantum circuit that will run on the quantum hardware, mapping the problem into qubits and gates. Once the quantum job completes, its output is fed back into the classical simulation pipeline, which aggregates and visualizes the results in a form domain experts can interpret.

IBM positions quantum not as a separate, exotic system but as another specialized accelerator in the HPC stack, similar to how GPUs and FPGAs are used today. Over roughly three decades, HPC workflows have evolved from CPU-only to heterogeneous architectures; quantum processors are presented here as the next accelerator class for problems like many-body physics, quantum chemistry and complex optimization that are difficult to scale classically. This interview, filmed at SC25 in St Louis, frames that shift as an incremental step in established HPC practice rather than a replacement of existing compute.

On the software side, the key piece is the integration between IBM Spectrum LSF and Qiskit, IBM’s open-source Python SDK for building and running quantum circuits on real hardware or simulators LSF submits, queues and schedules the classical pre- and post-processing jobs, and at the appropriate point in the workflow it invokes the quantum step via Qiskit APIs against IBM Quantum systems hosted in the same cloud region. This provides a single workload management layer and unified control plane for both classical MPI-style jobs and quantum execution.

The conversation also touches on accessibility: researchers and developers can already request limited free access to IBM Quantum systems via the IBM Cloud interface, experiment with Qiskit to prototype algorithms, simulate them, and then run them on real devices to study behavior under noise. The demo in the booth is less about a one-off visualization and more about showing how a complete hybrid pipeline can be expressed as jobs, scheduled, monitored and iterated as quantum hardware and cloud software continue to mature.

I’m publishing about 90+ videos from Embedded World North America 2025, I upload about 4 videos per day at 5AM/11AM/5PM/11PM CET/EST. Join https://www.youtube.com/charbax/join for Early Access to all 90 videos (once they’re all queued in next few days) Check out all my Embedded World North America videos in my Embedded World playlist here: https://www.youtube.com/playlist?list=PL7xXqJFxvYvjgUpdNMBkGzEWU6YVxR8Ga

This video was filmed using the DJI Pocket 3 ($669 at https://amzn.to/4aMpKIC using the dual wireless DJI Mic 2 microphones with the DJI lapel microphone https://amzn.to/3XIj3l8 ), watch all my DJI Pocket 3 videos here https://www.youtube.com/playlist?list=PL7xXqJFxvYvhDlWIAxm_pR9dp7ArSkhKK

Click the “Super Thanks” button below the video to send a highlighted comment under the video! Brands I film are welcome to support my work in this way 😁

Check out my video with Daylight Computer about their revolutionary Sunlight Readable Transflective LCD Display for Healthy Learning: https://www.youtube.com/watch?v=U98RuxkFDYY

source https://www.youtube.com/watch?v=Gbl-smbCW4g

Murata showcases its latest industrial power and sensing portfolio, from predictive maintenance vibration sensors to compact DC-DC converters, PoE modules and data center supplies. The video walks through how these building blocks support robotics, medical devices and GPU servers by combining isolation, efficiency and protection features. More technical details on Murata’s DC-DC converter families are available at https://www.murata.com/products/power/nonisolated-dc-dc-converter

On the SPS Nuremberg 2025 show floor, the demo starts with a predictive maintenance setup where a small vibration sensor on a rotating shaft streams condition-monitoring data in real time. By analyzing vibration signatures and trends, engineers can detect bearing wear, imbalance or misalignment early instead of waiting for unplanned downtime. The same sensing know-how underpins Murata’s broader industrial IoT strategy, feeding edge gateways, dashboards and maintenance workflows that close the loop from raw data to actionable insight downstream.

The tour then covers communication and power modules for OT networks. PoE DC-DC modules deliver up to 72 W over Ethernet to cameras, access points and edge nodes, while non-isolated converters and quarter-brick modules cover motor drives and PLC backplanes. Ultra-thin charge-pump modules reach around 72 W at close to 97% peak efficiency in only 2.1 mm height to convert 48 V to a 12 V intermediate bus in datacenters and 5G infrastructure. In robotics, dedicated converters, motor-drive rails and gate-drive supplies are tuned for fast transients and thermal limits inside collaborative and humanoid robot platforms on the factory floor.

A key highlight is Murata’s focus on data center and GPU power delivery. The interview shows 3×5″ 600 W AC-DC units, larger AC-DC platforms with around 95% efficiency, and a 2 kW GPU power concept that uses dual 1 kW non-isolated converters and high-current point-of-load stages to bring 48 V down to around 1 V rails. This two-stage approach mirrors the industry move to 48 V racks with intermediate bus and local high-current regulators for accelerators, storage and networking equipment.

The conversation closes by zooming out to Murata’s wider catalog, from MLCCs and high-voltage resistors for medical and industrial equipment to RFID, wireless connectivity and sensor lines shipping at scale. Viewers get a compact overview of how Murata’s passive components, sensing devices and power modules interlock – from predictive maintenance at the edge to high-density power shelves in the data center – offering a coherent toolbox for modern industrial and compute power workflow.

This video was filmed using the DJI Pocket 3 ($669 at https://amzn.to/4aMpKIC using the dual wireless DJI Mic 2 microphones with the DJI lapel microphone https://amzn.to/3XIj3l8 ), watch all my DJI Pocket 3 videos here https://www.youtube.com/playlist?list=PL7xXqJFxvYvhDlWIAxm_pR9dp7ArSkhKK

Click the “Super Thanks” button below the video to send a highlighted comment under the video! Brands I film are welcome to support my work in this way 😁

Check out my video with Daylight Computer about their revolutionary Sunlight Readable Transflective LCD Display for Healthy Learning: https://www.youtube.com/watch?v=U98RuxkFDYY

source https://www.youtube.com/watch?v=FmsR0QJYjIM

eNeural Technologies showcases eSL-Craft, a self-learning visual AI pipeline designed to keep edge models continuously up to date without manual retraining. Built on Advantech’s WA AIoT software platform and H-box edge systems, the service automates data collection, model retraining and deployment so inspection, logistics and robotics workloads can adapt to new conditions directly at the edge pipeline. https://www.eneural.ai/

At the core of the demo is eSL-Craft, a patented “adaptive self-learning” system that can refine object detection and segmentation models on real production data with minimal human labeling. It is part of eNeural’s broader AI-Craft toolchain, which combines automated annotation, model architecture optimization and quantization to shrink edge models while preserving accuracy and cutting AI time-to-market by as much as six times, particularly on NPU and embedded vision platforms using compact convolutional networks and mixed-precision training.

In this SPS 2025 demo at the Advantech booth in Nuremberg, a camera plus H-box edge GPU tracks pallets in a warehouse scenario and detects “corner cases” when new objects or packaging types appear. Instead of exporting raw video to the cloud, the system logs only the relevant samples, schedules on-device retraining on a zonal master node and pushes updated models back to the local inference devices, creating a closed loop between perception, data selection and continuous model refinement inside the factory line.

The architecture follows an edge-first and zone-based pattern: multiple cameras and NPU-powered clients handle real-time inference, while a GPU-equipped master node in each zone performs periodic retraining and then synchronizes only model weights with a central server. This resembles federated continual learning in industrial environments, keeping sensitive imagery on premises while still aggregating model improvements globally for robust multi-site deployment.

Looking ahead, the team discusses extending the same pipeline to mobile and humanoid robots that observe people and environments during the day, then retrain locally while charging at night to learn new faces, layouts and behaviors without exposing raw personal data. Today the heavy training runs on GPUs, with NPUs dedicated to low-latency inference, but the roadmap clearly targets broader use across smart factories, warehouses and service robotics, delivered as a combined Advantech hardware plus eNeural self-learning AI subscription for industrial customers.

source https://www.youtube.com/watch?v=JYkPcelILeQ

Finder shows how it is evolving from a relay specialist into a full control-cabinet and industrial automation partner, using its Opta micro PLC and 7M smart energy meters to connect classic switching hardware with data-driven control. The video walks through use cases such as lidar-based people detection, compact PLC control and cabinet-level power management, illustrating a path from simple relay logic to secure Industrial IoT architectures. https://opta.findernet.com/en/

At the center is the Opta PLC, co-developed with Arduino as a secure micro PLC based on a dual-core STM32H747 Arm Cortex-M7/M4 MCU and programmable both in C++/Arduino sketches and IEC 61131-3 languages like Ladder and Function Block Diagram. Offered in Ethernet, RS-485 and Wi-Fi/BLE variants, and expandable to dozens of I/Os, Opta can handle lidar distance sensing up to about two meters, machine control and edge data acquisition. A built-in crypto device enables authenticated, encrypted communications so the same hardware can scale from stand-alone logic control to connected PLC nodes.

Finder then positions its broader control-cabinet portfolio around Opta: interface and safety relays, DIN-rail power supplies from a few amps up to more than 40 A, and accessories for cabinet heating, ventilation and service lighting. A magnetic LED lamp that can run from the panel or on battery supports maintenance when cabinets are de-energized, and a range of panel sockets (Italian, French, US and both Swiss standards) lets panel builders export identical designs worldwide. Filmed at SPS 2025 in Nuremberg, the booth highlights this modular, export-ready cabinet platform.

Energy management is the second pillar. MID-certified 7M energy meters for single-phase and three-phase loads up to 80 A provide measurements that can be read and configured via NFC with the Finder Toolbox NFC smartphone app, even when the meter is not powered. Through Modbus or M-Bus these meters stream data into Opta, enabling dashboards that compare production, office consumption and EV charging, and supporting load-shedding or tariff-driven control. This metering layer is presented as a basis for Industry 5.0 incentives and ESG reporting, where verifiable energy data is becoming mandatory.

The interview closes on Finder’s ecosystem approach. Opta and the 7M/6M metering families are positioned as platforms for system integrators and OEMs, with Finder focusing on hardware, Arduino and IEC-61131-3 compatibility and cybersecurity features rather than vertical application software. A growing network of partners is expected to build sector-specific applications on top, from control panels to brownfield energy retrofits, using familiar cabinet components instead of proprietary black-box systems.

source https://www.youtube.com/watch?v=F1GLKf8LdxA

Xentara is a software-defined automation platform that replaces traditional hardware PLCs with a microservice-based realtime control layer running on standard industrial PCs. It allows engineers to implement deterministic control logic in C++ or Rust, run it on Linux with preempt-rt, and expose all process data via modern fieldbuses and open APIs, blending IT and OT in one control stack. https://xentara.io/

In this video, Xentara is shown powering a BenThor laser marking cell where a classic PLC has been completely removed and replaced by the software platform running on an Advantech IPC. The BENTHORcube concept focuses on compact, flexible laser systems for marking and engraving, and here Xentara provides the realtime execution environment, I/O handling and safety logic while the machine performs QR-code marking on small “business card” blanks. The demo is part of the joint Xentara, Advantech, Hilscher and BenThor presence at SPS 2025 in Nuremberg, illustrating how a standardized software layer can be reused across many different machine variants in one ecosystem.

Engineering is done with Xentara Workbench, where system integrators configure EtherCAT fieldbus, OPC UA server endpoints, semantic data models, timing and schedulers, and fine-grained access control lists in one environment. The semantic data layer allows the machine to be modeled as a digital representation with typed signals, states and relationships instead of ad-hoc tags, making it easier to plug that model into MES, cloud analytics or digital twin tools. Combined with secure OPC UA exposure of OT data to higher-level systems, the platform is designed to keep classic control loops deterministic while still fitting into a modern data-driven workflow.

On top of realtime control, Xentara is positioning itself as an IT-OT convergence layer ready for AI and agentic workloads. The team is working on an MCP (Model Context Protocol) interface so that large language models and other AI agents can safely access machine semantics and live data as standardized tools, while partners plug in their own ML models via ONNX and similar runtimes. This gives machine builders a path toward AI-assisted engineering, automated parameter optimization and natural-language interaction with production assets without turning the control platform itself into a monolithic AI product in the market.

The laser cell shown at the Advantech booth combines a JavaScript-based HMI, a robotics interface, websockets and the C++ control logic into one coherent microservice architecture orchestrated by Xentara. Operators simply insert a part, select a job and let the system execute a fully synchronized cycle that includes safety interlocks, motion, laser firing and QR-code generation, while the same architecture could be reused for cutting, molding or other processes. This makes the demo a concrete example of how software-defined automation, semantic modeling and early AI integration can be layered onto existing industrial hardware to evolve machines beyond the classic PLC-centric demo.

Xentara IT-OT convergence with Benthorcube laser cell, C++ and Rust control, MCP, OPC UA, EtherCAT
Xentara SPS laser marking cell with Advantech IPC, microservices PLC replacement, OPC UA, MCP, AI

source https://www.youtube.com/watch?v=AQ5A6WLQOBs

Handshake Public is a services and AI-enabled tendering partner for companies that want to sell to the public sector without building an in-house bid team. Instead of navigating fragmented portals, dense RFPs and strict EU public procurement rules alone, vendors outsource the full tender lifecycle: opportunity scouting, qualification, document preparation, submission and post-award administration. The business model is success-based: they prepare and manage bids across the EU and other markets, and clients only pay when a contract is actually won, with an initial “first win” offer for new customers. https://www.handshakepublic.com/

—
HDMI® Technology is the foundation for the worldwide ecosystem of HDMI-connected devices; integrated with displays, set-top boxes, laptops, audio video receivers and other product types. Because of this global usage, manufacturers, resellers, integrators and consumers must be assured that their HDMI® products work seamlessly together and deliver the best possible performance by sourcing products from licensed HDMI Adopters or authorized resellers. For HDMI Cables, consumers can look for the official HDMI® Cable Certification Labels on packaging. Innovation continues with the latest HDMI 2.2 Specification that supports higher 96Gbps bandwidth and next-gen HDMI Fixed Rate Link technology to provide optimal audio and video for a wide range of device applications. Higher resolutions and refresh rates are supported, including up to 12K@120 and 16K@60. Additionally, more high-quality options are supported, including uncompressed full chroma formats such as 8K@60/4:4:4 and 4K@240/4:4:4 at 10-bit and 12-bit color.
—

In the interview, founder and CEO Andrew Fowler explains how Handshake Public targets the multi-trillion public procurement market where every purchase, from automotive fleets and medical devices to legal services and IT consulting, must go through regulated tendering. Their platform combines deal sourcing, competitive analysis, smart monitoring and price optimization to track open tenders across jurisdictions, map them to a client’s products and then handle the detailed response work. That includes drafting and assembling technical annexes, compliance matrices, qualification forms and commercial schedules so vendors can focus on delivery rather than bureaucracy.

Filmed at Web Summit Lisbon 2025, the discussion focuses on how this approach particularly benefits startups and SMEs with strong private-sector revenue but little or no public-sector track record. Handshake Public uses automated monitoring to scan public portals, then filters out tenders whose “weasel words” and eligibility clauses would exclude inexperienced suppliers, helping companies avoid wasting resources on unwinnable RFPs. For viable opportunities, they structure responses to align with EU procurement directives and local procedures while ensuring that every formal requirement, from CPV codes to mandatory certifications, is properly addressed.

Fowler also touches on the long-term nature of public contracts and why a pay-per-win model aligns incentives between the service provider and vendors. Many framework agreements and multi-year contracts demand ongoing communication with contracting authorities, including invoicing rules, reporting obligations and contract change notices; Handshake Public remains involved in this post-award administration as part of an end-to-end public sector sales channel. The result is a hybrid of consultancy and SaaS-enabled operations for companies that want EU and US public sector customers in their revenue mix without becoming procurement experts themselves.

I’m publishing about 90+ videos from Embedded World North America 2025, I upload about 4 videos per day at 5AM/11AM/5PM/11PM CET/EST. Join https://www.youtube.com/charbax/join for Early Access to all 90 videos (once they’re all queued in next few days) Check out all my Embedded World North America videos in my Embedded World playlist here: https://www.youtube.com/playlist?list=PL7xXqJFxvYvjgUpdNMBkGzEWU6YVxR8Ga

This video was filmed using the DJI Pocket 3 ($669 at https://amzn.to/4aMpKIC using the dual wireless DJI Mic 2 microphones with the DJI lapel microphone https://amzn.to/3XIj3l8 ), watch all my DJI Pocket 3 videos here https://www.youtube.com/playlist?list=PL7xXqJFxvYvhDlWIAxm_pR9dp7ArSkhKK

Click the “Super Thanks” button below the video to send a highlighted comment under the video! Brands I film are welcome to support my work in this way 😁

Check out my video with Daylight Computer about their revolutionary Sunlight Readable Transflective LCD Display for Healthy Learning: https://www.youtube.com/watch?v=U98RuxkFDYY

source https://www.youtube.com/watch?v=ePeMx4mZq8Q

Rittal North America uses this booth tour to show why AI data centers and hyperscale GPU clusters depend on robust physical infrastructure, not just servers and software. The focus is on modular enclosures, Open Compute Project (OCP) racks, DC busbar power distribution and single-phase direct liquid cooling that can remove over one megawatt of thermal load from dense IT footprints, enabling deployment of large language model training and other compute-intensive workloads in production data infrastructure. https://www.rittal.com/com-en/products/Innovations/Direct-Liquid-Cooling

—
HDMI® Technology is the foundation for the worldwide ecosystem of HDMI-connected devices; integrated with displays, set-top boxes, laptops, audio video receivers and other product types. Because of this global usage, manufacturers, resellers, integrators and consumers must be assured that their HDMI® products work seamlessly together and deliver the best possible performance by sourcing products from licensed HDMI Adopters or authorized resellers. For HDMI Cables, consumers can look for the official HDMI® Cable Certification Labels on packaging. Innovation continues with the latest HDMI 2.2 Specification that supports higher 96Gbps bandwidth and next-gen HDMI Fixed Rate Link technology to provide optimal audio and video for a wide range of device applications. Higher resolutions and refresh rates are supported, including up to 12K@120 and 16K@60. Additionally, more high-quality options are supported, including uncompressed full chroma formats such as 8K@60/4:4:4 and 4K@240/4:4:4 at 10-bit and 12-bit color.
—

At the heart of the demo is Rittal’s rack-format Coolant Distribution Unit, a water-based direct-to-chip liquid cooling (DLC) system designed around OCP ORV3 form factors and GPU sleds. Each drawer is rated for roughly 250 kW of heat removal capacity, so operators can deploy 250, 500, 750 or 1 MW per row with an N+1 spare module for service and redundancy. Technology water circulates through supply and return manifolds to cold plates at the processor level, while facility water on the secondary loop carries that heat back toward chiller plants and external heat rejection equipment across the site.

The conversation dives into deployment topics such as growth planning, hydraulic design and power provisioning. A single megawatt CDU can require on the order of 400 gallons per minute of flow, so a 20 MW campus implies several thousand gallons per minute together with matching electrical distribution and grid capacity. Rittal stresses modularity so operators can start with a few OCP racks at 250 or 500 kW of cooling and scale out row by row as AI demand ramps instead of overbuilding a full megawatt cluster on day one.

Beyond direct-to-chip DLC, the video also covers Rittal Liquid Cooling Package (LCP) units that provide roughly 50–60 kW of air-to-water cooling for conventional 19-inch rack deployments. These in-row LCP systems pull hot exhaust air from standard server, storage and networking racks, pass it through coil heat exchangers and return chilled air to the IT aisle, making them suitable for mixed environments where traditional servers coexist with liquid-cooled OCP racks in the same white space. This flexibility lets operators handle both legacy and next-generation workloads in one coherent thermal architecture.

Filmed at Supercomputing 2025 in St Louis, the discussion closes on monitoring and control, with sensors, leak detection and networked logic that allow data center teams to manage coolant distribution and thermal headroom from a central operations console. Instead of simply moving water, the CDU and LCP systems dynamically allocate flow to the hottest cabinets, maintain redundancy at the pump level and support gradual migration from air cooling to liquid cooling as AI, HPC and cloud workloads continue to grow across the industry.

I’m publishing about 90+ videos from Embedded World North America 2025, I upload about 4 videos per day at 5AM/11AM/5PM/11PM CET/EST. Join https://www.youtube.com/charbax/join for Early Access to all 90 videos (once they’re all queued in next few days) Check out all my Embedded World North America videos in my Embedded World playlist here: https://www.youtube.com/playlist?list=PL7xXqJFxvYvjgUpdNMBkGzEWU6YVxR8Ga

This video was filmed using the DJI Pocket 3 ($669 at https://amzn.to/4aMpKIC using the dual wireless DJI Mic 2 microphones with the DJI lapel microphone https://amzn.to/3XIj3l8 ), watch all my DJI Pocket 3 videos here https://www.youtube.com/playlist?list=PL7xXqJFxvYvhDlWIAxm_pR9dp7ArSkhKK

Click the “Super Thanks” button below the video to send a highlighted comment under the video! Brands I film are welcome to support my work in this way 😁

Check out my video with Daylight Computer about their revolutionary Sunlight Readable Transflective LCD Display for Healthy Learning: https://www.youtube.com/watch?v=U98RuxkFDYY

source https://www.youtube.com/watch?v=8Ssp1t-g_wM

Exor International and its software company Corvina show how they combine industrial HMIs, edge gateways and an industrial IoT cloud to digitalize factory operations and OEM machines. On a real test rig from Exor’s Verona factory, the MicroEdge gateway, XPLC soft PLC (IEC 61131-3) and Corvina platform control a display test line, collect data and expose everything through browser-based dashboards and JMobile visualizations. https://www.exorint.com/products/hardware/series/microedge

On the shopfloor side, an Exor HMI with an integrated camera runs an AI-based access-control and operator-tracking algorithm, performing face recognition and user role management directly at the industrial edge. The same HMI executes PLC logic via XPLC and Dockerized containers on the MicroEdge hardware, blending deterministic control with containerized IIoT workloads. This lets OEMs deploy human–machine interfaces, soft PLCs and AI workloads on ARM or Intel-based edge devices, depending on performance and application requirements on site.

Corvina’s X Platform then aggregates machine data into real-time and historical analytics, turning the test machine into a digital twin with KPIs such as OEE and energy consumption. No-code dashboards let engineers build cloud or edge visualizations using OPC UA, MQTT and database connectors, while AI helps generate after-sales documentation and support flows. Filmed at the SPS 2025 show in Nuremberg, the demo reflects how Corvina is used in smart manufacturing and smart machinery deployments worldwide.

Around the booth, Exor presents a hardware portfolio built on the same software stack: IP69 food-and-beverage HMIs for washdown environments, outdoor sunlight-readable displays, marine bridge panels, X5 wireless safety handhelds with SIL3-rated emergency functions, and rugged “field HMI” tablets with Wi-Fi, NFC and integrated cameras. Industrial PCs and open HMIs span ARM and x86 architectures, from MicroEdge Basic and MicroEdge Plus gateways up to x86-based eXware and eX200 “edge HMI” systems, all capable of running JMobile and acting as IIoT gateways for Corvina.

Cybersecurity and regulatory alignment are central: MicroEdge acts as a secure router and VPN & IIoT gateway, with firewalls, network segmentation and IEC 62443-ready architectures to help customers address CRA and NIS2 requirements. Through Corvina, fleets of machines can be patched and updated over-the-air, vulnerabilities (CVEs) tracked, and new applications rolled out at scale. Exor and Corvina organize their business around smart manufacturing and smart machinery segments, enabling factories and machine builders worldwide to move from pure hardware sales to data-driven services, remote monitoring and pay-per-use as a service offering.

This video was filmed using the DJI Pocket 3 ($669 at https://amzn.to/4aMpKIC using the dual wireless DJI Mic 2 microphones with the DJI lapel microphone https://amzn.to/3XIj3l8 ), watch all my DJI Pocket 3 videos here https://www.youtube.com/playlist?list=PL7xXqJFxvYvhDlWIAxm_pR9dp7ArSkhKK

Click the “Super Thanks” button below the video to send a highlighted comment under the video! Brands I film are welcome to support my work in this way 😁

Check out my video with Daylight Computer about their revolutionary Sunlight Readable Transflective LCD Display for Healthy Learning: https://www.youtube.com/watch?v=U98RuxkFDYY

source https://www.youtube.com/watch?v=6A_qJjdB-eA

Chiahung Hung, Product Sales Manager for @AdvantechCorp Europe’s IoT Automation sector, explains how @Codesys-AutomationSoftware turns standard industrial PCs into flexible, software-defined controllers that break the traditional proprietary PLC lock-in. By running the CODESYS runtime on x86 or ARM-based IPCs, Advantech can offer “soft PLC” functionality across its AMAX and MX controller families, letting machine builders decouple application logic from specific hardware platforms and adopt an Android-like model for industrial control. This opens the door to using the same IEC 61131-3 project on multiple devices, from compact edge controllers to panel PCs, with lifecycle managed as software rather than as fixed hardware generations, and more details can be found on the Advantech website https://www.advantech.com/

At the heart of the demo is an AMAX-5570 Atom-based edge controller, shown driving third-party servo motors in tightly synchronized motion control. The platform is optimized for CODESYS with a 1 ms real-time environment, supporting up to 32-axis motion control with 500 µs EtherCAT cycle times under Linux, plus SoftMotion and CNC/robotics libraries for more complex kinematics. The classic “pencil on two spinning shafts” test illustrates deterministic behavior under high speed, showing that even an entry-level controller can handle precise multi-axis coordination while remaining cost-efficient in the field, which makes it attractive as a general-purpose automation setup.

Beyond this single controller, Hung positions AMAX and MX as a complete CODESYS-ready portfolio, including UNO box PCs and AMAX panel controllers that combine HMI and control in one device with web-based visualization. In the video, an AMAX panel unit runs CODESYS Web Visualization as an HMI, reading live data from the UNO controller and rendering dashboards directly in a browser-like interface, which simplifies deployment of machine and factory UIs. Recorded at SPS 2025 in Nuremberg, the setup reflects Advantech’s broader focus at the show on open automation, Edge AI-ready hardware, and IT/OT converged architectures built around software-defined control.

A key step this year is that Advantech’s CODESYS platforms no longer target only Windows; Linux runtimes are now supported across selected AMAX controllers, with more SKUs being qualified as “CODESYS Enabled” over time. This Linux support matters for customers standardizing on containerized edge workloads and long-term LTS distributions in OT environments, where they can align PLC workloads with other edge services running on the same hardware. Hung also points out that Advantech is working toward full EU Cyber Resilience Act readiness across its controller portfolio by 2026, addressing concerns from OEMs who worry that embedding a PC-class platform inside machines might increase the attack surface, by integrating secure boot, TPM, and CRA-aligned hardening as part of the standard device compliance.

For machine builders, system integrators, and OT engineers, the message is that automation is becoming a software-defined layer running on standardized compute, not a collection of closed PLC islands. With CODESYS as the common runtime, Advantech edge controllers can scale from Intel Atom-based AMAX-5570 V2 units up to Core-powered panel controllers such as AMAX-PT800, while offering industrial Ethernet and fieldbus connectivity via EtherCAT, PROFINET, Ethernet/IP, CANopen, OPC UA, and Modbus TCP/RTU for both motion and process applications. This gives factories and infrastructure operators a path to adopt newer edge computing and AI capabilities without discarding existing engineering practices, effectively treating the control layer as a portable software asset instead of a fixed hardware portfolio.

I’m publishing about 50+ videos from SPS Nuremberg 2025 check playlist https://www.youtube.com/playlist?list=PL7xXqJFxvYvgeW6Uf8MIlo7hpxuysbiCc I upload about 4 videos per day at 5AM/11AM/5PM/11PM CET/EST (from other recent events too)

source https://www.youtube.com/watch?v=K2_RghrbtG8

ctrlX OS from Bosch Rexroth is presented here running on Advantech’s UNO edge IPCs as an app-based automation platform built on a real-time capable Linux with Ubuntu Core and snap packages. The idea is “smartphone of automation”: you pick the industrial PC form factor you want from the UNO family, get a pre-validated OS image with core apps like firewall and VPN, and then extend it with your own apps from an ecosystem rather than rebuilding the stack each time. This gives OEMs, machine builders and IT/OT teams a consistent software base across very different hardware profiles. https://www.ctrlx-os.com/en/ctrlx-os/

In the demo, ctrlX WORKS shows both a Bosch Rexroth ctrlX CORE controller and an Advantech UNO device side by side, all managed through the same web UI and REST APIs. The platform is explicitly API-first: every setting exposed in the UI, from firewall rules to user management, can be scripted via REST, so one script can roll out configuration to a single device or to fleets of hundreds. Apps such as an IoT dashboard based on Grafana, Node-RED with OPC UA connectors and a Python runtime are installed, updated or removed with one click or via automated scripts, aligning industrial automation with DevOps practices.

A key concept is the ctrlX Data Layer, which acts as a structured, real-time namespace for everything the device knows: CPU metrics, field I/O, application state and custom variables. That data tree can be traversed, filtered and mirrored between devices so that one node can be the “hero” and others the “sidekicks,” all sharing a synchronized data model. This makes it easier to build distributed control and analytics topologies where edge IPCs, controllers and higher-level IT systems share a common view of the plant without custom glue code for each box.

For developers, ctrlX OS behaves like a modern software platform rather than a closed PLC. Bosch Rexroth provides an SDK, documentation and reference apps on GitHub, plus an app-development guideline so partners can package their own snaps that integrate cleanly with the OS, data layer and security model. The ctrlX OS Store then acts as a digital marketplace where OEMs, ISVs and integrators publish apps ranging from connectivity and security to HMI and AI, and customers can add them to any ctrlX OS-based device with standard licensing. Because the OS is hardware-independent and available as a virtual machine, the same app portfolio can span controllers, edge IPCs, HMIs and on-prem or cloud infrastructure.

Cybersecurity and regulation are central to the story. ctrlX OS is positioned as “secure by design” and “secure by default,” certified to IEC 62443-4-2 Security Level 2 and prepared for Europe’s Cyber Resilience Act, including a structured process for vulnerability management and security patch delivery during operation. Built-in functions such as certificate handling, role-based access with LDAP integration, encrypted backups and firewall/VPN apps aim to make it realistic for OT teams to maintain a secure posture over the full lifecycle without building their own Linux distribution.

Recorded at the Advantech booth during SPS – Smart Production Solutions in Nuremberg, the discussion also highlights the OEM partnership between Bosch Rexroth and Advantech: UNO edge IPCs ship from the factory with ctrlX OS and a curated set of apps, turning them into ready-to-use edge gateways and controllers that can then be tailored per project through additional apps and scripts. The commercial model is deliberately flexible: you buy hardware with a base OS image and partners can choose subscription or perpetual licensing per app, while integrators can bundle their own apps on top as part of full project deliveries. The result is a sector-agnostic platform for energy, manufacturing, infrastructure or any other domain where developers want to focus on domain logic and let an open industrial Linux handle real-time execution, fleet management and security.

source https://www.youtube.com/watch?v=2KuMPlkhvyU

Tesyda is a Prague-based SaaS startup building BESEGMA, an AI-powered predictive analytics platform that turns historical business data into forecasts for revenue, churn and customer behaviour. The platform combines machine learning models and behavioural segmentation to generate propensity scores, highlight high-value leads and surface risks across the customer lifecycle, all exposed through a business-friendly interface rather than raw data science tooling.

Besegma

—
HDMI® Technology is the foundation for the worldwide ecosystem of HDMI-connected devices; integrated with displays, set-top boxes, laptops, audio video receivers and other product types. Because of this global usage, manufacturers, resellers, integrators and consumers must be assured that their HDMI® products work seamlessly together and deliver the best possible performance by sourcing products from licensed HDMI Adopters or authorized resellers. For HDMI Cables, consumers can look for the official HDMI® Cable Certification Labels on packaging. Innovation continues with the latest HDMI 2.2 Specification that supports higher 96Gbps bandwidth and next-gen HDMI Fixed Rate Link technology to provide optimal audio and video for a wide range of device applications. Higher resolutions and refresh rates are supported, including up to 12K@120 and 16K@60. Additionally, more high-quality options are supported, including uncompressed full chroma formats such as 8K@60/4:4:4 and 4K@240/4:4:4 at 10-bit and 12-bit color.
—

In the demo, Monika shows how non-technical users can work with an embedded copilot inside BESEGMA: you describe what you want to predict, such as next year’s turnover or customer churn, and the system builds a model on top of your historical CRM, ERP or Excel/CSV data. Tesyda provides native connectors to common CRM systems and can add custom connectors, so data ingestion becomes a one-click part of the workflow rather than an IT project. Once trained on your own historical records, the platform outputs dashboards, graphs and exports to Excel that managers can plug directly into their planning workflow.

Beyond the interface, Tesyda emphasises data privacy and control. BESEGMA runs as a standalone cloud platform with in-house AI models, avoiding external AI APIs so that propensity models and prediction outputs never leave Tesyda’s controlled environment. The stack includes built-in encryption, data anonymisation and GDPR-compliant hosting, with a design where even Tesyda’s team cannot access raw customer data unless explicitly requested for support, which is important for regulated sectors such as pharmaceuticals and manufacturing.

Recorded at Web Summit Lisbon 2025, the conversation also touches on how the product emerged from Monika’s previous experience as CEO of a telecom and cloud software company, where large volumes of operational and sales data were collected but rarely converted into actionable forecasts. Tesyda’s team of roughly ten people, mostly developers, is now focused on extending BESEGMA’s copilot capabilities, adding more CRM and ERP integrations and refining the forecasting engine so that small and mid-sized companies can use the same techniques as larger data teams without needing in-house data scientists.

Commercially, BESEGMA is offered on a subscription basis, typically in the low hundreds of euros per month, with a built-in allowance of predictions and the option to top up usage when running more scenarios. Under the hood, the platform implements Behavioural Segmentation Analysis and custom propensity models tailored to each customer, helping users prioritise leads, shorten deal cycles and test what-if scenarios around churn or demand without pretending to be 100% certain about the future.

I’m publishing about 90+ videos from Embedded World North America 2025, I upload about 4 videos per day at 5AM/11AM/5PM/11PM CET/EST. Join https://www.youtube.com/charbax/join for Early Access to all 90 videos (once they’re all queued in next few days) Check out all my Embedded World North America videos in my Embedded World playlist here: https://www.youtube.com/playlist?list=PL7xXqJFxvYvjgUpdNMBkGzEWU6YVxR8Ga

This video was filmed using the DJI Pocket 3 ($669 at https://amzn.to/4aMpKIC using the dual wireless DJI Mic 2 microphones with the DJI lapel microphone https://amzn.to/3XIj3l8 ), watch all my DJI Pocket 3 videos here https://www.youtube.com/playlist?list=PL7xXqJFxvYvhDlWIAxm_pR9dp7ArSkhKK

Click the “Super Thanks” button below the video to send a highlighted comment under the video! Brands I film are welcome to support my work in this way 😁

Check out my video with Daylight Computer about their revolutionary Sunlight Readable Transflective LCD Display for Healthy Learning: https://www.youtube.com/watch?v=U98RuxkFDYY

source https://www.youtube.com/watch?v=7vGxIe68gYE

Advantech’s European IoT leadership comes through clearly in this conversation with Jash Bansidhar, AVP of Industrial IoT sector for Advantech in Europe. The discussion frames Advantech as a global reference in industrial PCs (more than 40% market share) and industrial I/O, now extended with a broad stack that spans computing, communication, data acquisition and IoT software for edge-to-cloud integration, aligned with the corporate vision of “Enabling an Intelligent Planet.” The result is a hardware and software portfolio that lets OEMs, ISVs and integrators treat Advantech as a one-stop shop for modular, scalable industrial computing across many sectors. https://www.advantech.com/

A major focus in the video is next-generation control and I/O, where Advantech combines its own hardware heritage with key ecosystems such as CODESYS and Bosch Rexroth’s ctrlX OS. ctrlX OS brings Linux-based, app-centric automation with strong cybersecurity properties that map to the EU Cyber Resilience Act, while Advantech contributes UNO edge IPCs, industrial networking and I/O modules to build complete control architectures. The booth also highlights functional safety with virtualized controllers and safety I/O over PROFINET and EtherCAT, pointing toward PLC virtualization, app-based engineering and cyber-resilient plant architectures.

On top of the hardware, Advantech is pushing an IoT and edge software framework that connects OT devices to cloud-native AI pipelines. Their WISE-Edge / WISE-PaaS approach is about IT/OT convergence, normalizing data from controllers, gateways and IPCs and exposing it to AI developers through standardized APIs and SDKs. The wall of logos behind the interview—Intel, AMD, NVIDIA, NXP, Qualcomm, MediaTek, Rockchip and others—illustrates how the framework integrates silicon vendors’ AI stacks so ISVs can target different GPU and CPU platforms with a single architecture while experimenting with agentic AI workflows and agent builders for industrial use cases.

Jash also explains how Advantech organizes itself to support this at scale: local sales, technical and support teams in major European countries, complemented by channel and domain partners close to their customers, plus global R&D hubs in Taiwan, China and Europe. Vertical focus areas include manufacturing, energy and utilities, transportation, retail and healthcare, each with the necessary certifications such as IEC 61131 for PLCs and IEC 61850 for substations. Recorded at the SPS smart production show in Nuremberg, the interview situates Advantech within a wider ecosystem that includes Bosch Rexroth, SALZ Automation and others working on real-time data, predictive maintenance and connected factories.

The conversation closes on edge AI and physical AI, where Advantech is early to adopt NVIDIA Jetson Thor across products like the MIC-743 and dedicated robotics and medical AI platforms, delivering up to roughly 2,070 FP4 TFLOPS at the edge for perception, LLMs and multi-sensor fusion in compact systems. Combined with their automation, networking and IoT software stack, this positions Advantech to support emerging workloads such as humanoid robots on the factory floor, robotic cells, intelligent machines and distributed inferencing across brownfield and greenfield plants, all fed back into a joint silicon and system roadmap with key chip vendors for the coming years.

I’m publishing about 50+ videos from SPS Nuremberg 2025, I upload about 4 videos per day at 5AM/11AM/5PM/11PM CET/EST. Join https://www.youtube.com/charbax/join for Early Access to all 90 videos (once they’re all queued in next few days) Check out all my Embedded World North America videos in my Embedded World playlist here: https://www.youtube.com/playlist?list=PL7xXqJFxvYvjgUpdNMBkGzEWU6YVxR8Ga

This video was filmed using the DJI Pocket 3 ($669 at https://amzn.to/4aMpKIC using the dual wireless DJI Mic 2 microphones with the DJI lapel microphone https://amzn.to/3XIj3l8 ), watch all my DJI Pocket 3 videos here https://www.youtube.com/playlist?list=PL7xXqJFxvYvhDlWIAxm_pR9dp7ArSkhKK

Click the “Super Thanks” button below the video to send a highlighted comment under the video! Brands I film are welcome to support my work in this way 😁

Check out my video with Daylight Computer about their revolutionary Sunlight Readable Transflective LCD Display for Healthy Learning: https://www.youtube.com/watch?v=U98RuxkFDYY

source https://www.youtube.com/watch?v=k-VfpkZGW_U

Cerebras founder and chief architect Michael James walks through the CS-3 system and its wafer-scale engine, a single 300 mm die integrating around a million AI-optimized compute cores on one piece of silicon. Built in 5 nm with roughly 4 trillion transistors, WSE-3 delivers on-chip memory, interconnect and compute in one monolithic device, targeting high-throughput AI inference and data-intensive HPC workloads in a compact rack-scale node. https://www.cerebras.ai/chip

—
HDMI® Technology is the foundation for the worldwide ecosystem of HDMI-connected devices; integrated with displays, set-top boxes, laptops, audio video receivers and other product types. Because of this global usage, manufacturers, resellers, integrators and consumers must be assured that their HDMI® products work seamlessly together and deliver the best possible performance by sourcing products from licensed HDMI Adopters or authorized resellers. For HDMI Cables, consumers can look for the official HDMI® Cable Certification Labels on packaging. Innovation continues with the latest HDMI 2.2 Specification that supports higher 96Gbps bandwidth and next-gen HDMI Fixed Rate Link technology to provide optimal audio and video for a wide range of device applications. Higher resolutions and refresh rates are supported, including up to 12K@120 and 16K@60. Additionally, more high-quality options are supported, including uncompressed full chroma formats such as 8K@60/4:4:4 and 4K@240/4:4:4 at 10-bit and 12-bit color.
—

He explains the extreme power-delivery and packaging needed to run this chip at roughly 25 kW: front-side AC/DC modules, 3D power distribution and dense arrays of regulators positioned close to the wafer to manage around 30,000 amps of current. Because all compute and 44 GB of SRAM sit on a single wafer, the system minimizes off-chip traffic and uses control logic that smooths power ramps with dummy operations when workloads switch off, avoiding destructive current spikes while preserving energy efficiency.

On the architecture side, James describes the WSE-3 as a proprietary dataflow processor designed for strong scaling. Loop induction variables, data movement and network behavior are encoded directly into the instruction set, so a single matrix operation can be spread spatially across the full grid of cores with minimal software overhead. That allows Cerebras to map full transformer layers over the wafer and reach very high inference throughput, with customers reporting large speedups over Nvidia GPU clusters on latency-sensitive language-model serving.

The discussion then shifts to real workloads, including a global shallow-water-equation simulation of an asteroid impact off California, run at about 200 m resolution over the entire planet. By exploiting the dense on-wafer memory and mesh interconnect, a cluster of CS-3 nodes achieved exascale-class performance for this tsunami scenario at a fraction of the power draw of traditional exascale systems, while still supporting large language models such as Llama and DeepSeek on the same architecture.

Filmed at Supercomputing 2025 in St Louis, the interview also touches on manufacturing yield and roadmap. Cerebras overprovisions identical cores across the wafer and then uses automated defect mapping plus constraint solving to reroute communication around faulty regions, guaranteeing at least 900,000 working cores per device and turning the rest into pass-through fabric. James hints that future generations will continue this wafer-scale path, pushing AI inference and physics-based HPC further by co-designing architecture, packaging and dataflow software as a single system.

I’m publishing about 60+ videos Supercomputing 2025 #SC25 I upload about 4 videos per day at 5AM/11AM/5PM/11PM CET/EST. Join https://www.youtube.com/charbax/join for Early Access to all 90 videos (once they’re all queued in next few days) Check out all my Supercomputing 2025 SC25 videos in my playlist here: https://www.youtube.com/playlist?list=PL7xXqJFxvYvihnaq98TO55Cbe2VMD9mk8

This video was filmed using the DJI Pocket 3 ($669 at https://amzn.to/4aMpKIC using the dual wireless DJI Mic 2 microphones with the DJI lapel microphone https://amzn.to/3XIj3l8 ), watch all my DJI Pocket 3 videos here https://www.youtube.com/playlist?list=PL7xXqJFxvYvhDlWIAxm_pR9dp7ArSkhKK

Click the “Super Thanks” button below the video to send a highlighted comment under the video! Brands I film are welcome to support my work in this way 😁

Check out my video with Daylight Computer about their revolutionary Sunlight Readable Transflective LCD Display for Healthy Learning: https://www.youtube.com/watch?v=U98RuxkFDYY

source https://www.youtube.com/watch?v=tlvfZulPYaw

Supermicro uses this SC25 booth tour to show how its AI and HPC servers scale from NVIDIA HGX B300 Blackwell Ultra clusters to RTX PRO 6000 Blackwell Server Edition MGX systems and fully integrated liquid-cooled racks for enterprise AI factories and research data centers. The focus is on how 288GB HBM3e per GPU, 800Gb/s integrated networking and rack-scale thermal design let operators run large language models and high-density training or inference more efficiently than previous infrastructure. https://www.supermicro.com/en/accelerators/nvidia

—
HDMI® Technology is the foundation for the worldwide ecosystem of HDMI-connected devices; integrated with displays, set-top boxes, laptops, audio video receivers and other product types. Because of this global usage, manufacturers, resellers, integrators and consumers must be assured that their HDMI® products work seamlessly together and deliver the best possible performance by sourcing products from licensed HDMI Adopters or authorized resellers. For HDMI Cables, consumers can look for the official HDMI® Cable Certification Labels on packaging. Innovation continues with the latest HDMI 2.2 Specification that supports higher 96Gbps bandwidth and next-gen HDMI Fixed Rate Link technology to provide optimal audio and video for a wide range of device applications. Higher resolutions and refresh rates are supported, including up to 12K@120 and 16K@60. Additionally, more high-quality options are supported, including uncompressed full chroma formats such as 8K@60/4:4:4 and 4K@240/4:4:4 at 10-bit and 12-bit color.
—

On the Blackwell HGX side, Philip Tamaki walks through Supermicro’s multi-form-factor HGX B300 platforms, where eight B300 GPUs provide up to 2.3TB of HBM3e and are coupled with integrated ConnectX-8 SuperNICs that expose 800Gb/s OSFP ports per node, reducing earlier PCIe bottlenecks. Air-cooled and direct-liquid-cooled variants balance rack power density with CPU options such as AMD EPYC 9005 or Intel Xeon 6700, so clusters can be tuned to specific AI training or simulation workloads in practice.

The conversation then shifts to PCIe GPUs, where the NVIDIA RTX PRO 6000 Blackwell Server Edition with 96GB of GDDR7 and around 1.6TB/s of memory bandwidth makes PCIe-based servers attractive again for enterprise inference, retrieval-augmented generation and model serving. Supermicro is rolling out a broad portfolio of RTX PRO 6000 systems, including 4U MGX dual-root designs and compact edge nodes, so many production LLMs in the tens-of-billions of parameters can fit on a single GPU while additional GPUs handle vector search, orchestration and data processing.

At rack scale, the video highlights Supermicro’s NVIDIA GB300 NVL72 system, which combines 72 Blackwell GPUs and 36 Grace CPUs in a 48U rack with direct liquid cooling and NVLink domains up to 1.8TB/s, enabling very high AI performance in a single rack footprint. This ties into Supermicro’s Data Center Building Block Solutions strategy: pre-validated racks with in-rack cooling distribution units, vertical and horizontal manifolds, rear-door heat exchangers and sidecar CDUs that can dissipate 50–200kW of heat without requiring entirely new building infrastructure.

Finally, filmed at Supercomputing 2025 in St. Louis, the tour steps behind the cabinets to look at mock-up and real-world cooling towers, including the liquid-cooling tower deployed at Supermicro’s San Jose campus to validate closed-loop operation with production workloads. Alongside partner sessions from NVIDIA, AMD, Intel and software-defined storage vendors, the booth shows how Supermicro is positioning itself as a leading supplier of AI and HPC infrastructure by shipping complete Blackwell-era racks that shorten time-to-online for demanding customers.

I’m publishing about 90+ videos from Embedded World North America 2025, I upload about 4 videos per day at 5AM/11AM/5PM/11PM CET/EST. Join https://www.youtube.com/charbax/join for Early Access to all 90 videos (once they’re all queued in next few days) Check out all my Embedded World North America videos in my Embedded World playlist here: https://www.youtube.com/playlist?list=PL7xXqJFxvYvjgUpdNMBkGzEWU6YVxR8Ga

This video was filmed using the DJI Pocket 3 ($669 at https://amzn.to/4aMpKIC using the dual wireless DJI Mic 2 microphones with the DJI lapel microphone https://amzn.to/3XIj3l8 ), watch all my DJI Pocket 3 videos here https://www.youtube.com/playlist?list=PL7xXqJFxvYvhDlWIAxm_pR9dp7ArSkhKK

Click the “Super Thanks” button below the video to send a highlighted comment under the video! Brands I film are welcome to support my work in this way 😁

Check out my video with Daylight Computer about their revolutionary Sunlight Readable Transflective LCD Display for Healthy Learning: https://www.youtube.com/watch?v=U98RuxkFDYY

source https://www.youtube.com/watch?v=NTYJssgsfUQ

Seagate’s Exos 4U100 JBOD enclosure shown here is essentially a 4U storage brick that packs up to 100 Mozaic HAMR hard drives, reaching 3.2 petabytes in a single tray when populated with 32TB disks. It targets AI backends, data lakes and large backup repositories where cost per terabyte, rack density and power per petabyte really matter. https://www.seagate.com/products/storage/data-storage-systems/jbod/exos-4u74-and-4u100/

—
HDMI® Technology is the foundation for the worldwide ecosystem of HDMI-connected devices; integrated with displays, set-top boxes, laptops, audio video receivers and other product types. Because of this global usage, manufacturers, resellers, integrators and consumers must be assured that their HDMI® products work seamlessly together and deliver the best possible performance by sourcing products from licensed HDMI Adopters or authorized resellers. For HDMI Cables, consumers can look for the official HDMI® Cable Certification Labels on packaging. Innovation continues with the latest HDMI 2.2 Specification that supports higher 96Gbps bandwidth and next-gen HDMI Fixed Rate Link technology to provide optimal audio and video for a wide range of device applications. Higher resolutions and refresh rates are supported, including up to 12K@120 and 16K@60. Additionally, more high-quality options are supported, including uncompressed full chroma formats such as 8K@60/4:4:4 and 4K@240/4:4:4 at 10-bit and 12-bit color.
—

Torsten Konrad explains how the enclosure is engineered around power and cooling rather than just raw drive count. Front-to-back airflow is guided through the main chamber and also under the drive plane so every slot receives fresh air, yielding up to 70% better thermal efficiency and around 30% lower power than legacy enclosures. The system uses SAS-4 (24G SAS), delivering up to 24 GB/s aggregate bandwidth while still letting you treat it as a simple, high-density JBOD from the host point of view.

Filmed at Supercomputing 2025 (SC25) in St. Louis, the video also dives into practical deployment details that matter to data center operators. The Exos 4U100 is available in 1.0 m and 1.2 m chassis depths to match different rack standards, with internal rails so the whole unit slides onto a shelf without fiddly rail alignment. Internal cable management keeps SAS cabling inside the rack envelope, and because the system uses built-in carriers you can slot bare 3.5″ drives directly without extra hardware, simplifying service and drive swaps.

Logically it’s “just a bunch of disks,” but in modern architectures the JBOD becomes the capacity tier behind software-defined storage running on front-end servers. Erasure coding or RAID lives in the software layer, while this enclosure focuses on density, airflow, vibration damping and predictable thermals for 100 spinning disks. For AI workloads the first tier is typically HBM and NVMe flash feeding GPUs, with HDD-based tiers like Exos 4U100 holding cold and warm data, training corpora, model checkpoints and long-term logs at roughly one-fifth the cost per terabyte of flash.

Seagate’s Mozaic HAMR roadmap pushes capacity beyond today’s 32TB drives toward 40TB, 50TB and higher, and the 4U100/4U74 enclosures are explicitly designed to accommodate those future disks so petabytes per rack can keep scaling without changing the chassis. That makes a single 42U rack with up to ten of these trays a 32-petabyte building block for AI, edge and sovereign data infrastructures. Hardware root-of-trust, secure boot on the BMC and standards-based management interfaces are included to meet tightening EU and US regulatory requirements around storage security in large-scale environments.

I’m publishing about 90+ videos from Embedded World North America 2025, I upload about 4 videos per day at 5AM/11AM/5PM/11PM CET/EST. Join https://www.youtube.com/charbax/join for Early Access to all 90 videos (once they’re all queued in next few days) Check out all my Embedded World North America videos in my Embedded World playlist here: https://www.youtube.com/playlist?list=PL7xXqJFxvYvjgUpdNMBkGzEWU6YVxR8Ga

This video was filmed using the DJI Pocket 3 ($669 at https://amzn.to/4aMpKIC using the dual wireless DJI Mic 2 microphones with the DJI lapel microphone https://amzn.to/3XIj3l8 ), watch all my DJI Pocket 3 videos here https://www.youtube.com/playlist?list=PL7xXqJFxvYvhDlWIAxm_pR9dp7ArSkhKK

Click the “Super Thanks” button below the video to send a highlighted comment under the video! Brands I film are welcome to support my work in this way 😁

Check out my video with Daylight Computer about their revolutionary Sunlight Readable Transflective LCD Display for Healthy Learning: https://www.youtube.com/watch?v=U98RuxkFDYY

source https://www.youtube.com/watch?v=qZW7kCh2VDY

OpenMV CEO and co-founder Kwabena Agyeman walks through the new OpenMV AE3, a 1 inch by 1 inch edge-AI camera board designed to embed machine vision and neural networks into literally anything – from light switches to wearables. Built as a tiny, battery-friendly “Arduino of machine vision,” the AE3 combines MicroPython programmability with around 250 billion operations per second of NPU compute so developers can deploy real-time vision models without touching the cloud. https://openmv.io/products/openmv-ae3

—
HDMI® Technology is the foundation for the worldwide ecosystem of HDMI-connected devices; integrated with displays, set-top boxes, laptops, audio video receivers and other product types. Because of this global usage, manufacturers, resellers, integrators and consumers must be assured that their HDMI® products work seamlessly together and deliver the best possible performance by sourcing products from licensed HDMI Adopters or authorized resellers. For HDMI Cables, consumers can look for the official HDMI® Cable Certification Labels on packaging. Innovation continues with the latest HDMI 2.2 Specification that supports higher 96Gbps bandwidth and next-gen HDMI Fixed Rate Link technology to provide optimal audio and video for a wide range of device applications. Higher resolutions and refresh rates are supported, including up to 12K@120 and 16K@60. Additionally, more high-quality options are supported, including uncompressed full chroma formats such as 8K@60/4:4:4 and 4K@240/4:4:4 at 10-bit and 12-bit color.
—

Under the hood, the AE3 is powered by Alif Semiconductor’s Ensemble E3 SoC: dual Arm Cortex-M55 cores (400 MHz and 160 MHz) paired with two Ethos-U55 microNPUs delivering roughly 250 GOPS of AI throughput in that 6×7 mm package. On the board you get a 1 MP color global-shutter camera (up to ~120 fps at VGA), Octal-SPI flash streaming models at about 200 MB/s, integrated Wi-Fi/Bluetooth, high-speed USB-C, and a Qwiic connector for the SparkFun ecosystem. An onboard time-of-flight sensor, IMU (accelerometer/gyroscope), microphone and RGB LED turn the AE3 into a complete sensing and interaction platform rather than just a bare camera module.

In the demo, Kwabena shows the AE3 running hand-landmark detection at around 30 fps directly on the microcontroller, enabling gesture-driven user interfaces, as well as eye and face-landmark tracking, face recognition, segmentation and monocular depth estimation. The firmware stack supports a spectrum of neural network architectures – YOLOv2/v5/v8/v11 and custom CNNs – compressed to fit the constrained memory while still maintaining interactive frame rates. Because all inference runs locally, a smart door lock can recognize your face, or a smart light switch can see your hand, without ever streaming biometric data to a server.

He also explains how this board grew out of a decade of OpenMV work: eight years as a side project and the last two years full-time, culminating in a Kickstarter campaign for the N6 and AE3 that passed roughly $150k and around 700 backers. Filmed at Embedded World North America 2025, the conversation hints at where OpenMV is heading next – from battery-powered “AI in a golf cap” to industrial and robotics deployments – as MicroPython support for the Alif Ensemble family and the broader OpenMV ecosystem make it easier for engineers and makers to treat tiny AI vision as just another building block.

I’m publishing about 90+ videos from Embedded World North America 2025, I upload about 4 videos per day at 5AM/11AM/5PM/11PM CET/EST. Join https://www.youtube.com/charbax/join for Early Access to all 90 videos (once they’re all queued in next few days) Check out all my Embedded World North America videos in my Embedded World playlist here: https://www.youtube.com/playlist?list=PL7xXqJFxvYvjgUpdNMBkGzEWU6YVxR8Ga

This video was filmed using the DJI Pocket 3 ($669 at https://amzn.to/4aMpKIC using the dual wireless DJI Mic 2 microphones with the DJI lapel microphone https://amzn.to/3XIj3l8 ), watch all my DJI Pocket 3 videos here https://www.youtube.com/playlist?list=PL7xXqJFxvYvhDlWIAxm_pR9dp7ArSkhKK

Click the “Super Thanks” button below the video to send a highlighted comment under the video! Brands I film are welcome to support my work in this way 😁

Check out my video with Daylight Computer about their revolutionary Sunlight Readable Transflective LCD Display for Healthy Learning: https://www.youtube.com/watch?v=U98RuxkFDYY

source https://www.youtube.com/watch?v=vL-7HHBTUW0

Tirias Research founder and principal analyst Jim McGregor outlines how his firm connects semiconductor roadmaps, cloud infrastructure and emerging compute paradigms to guide technology vendors and enterprises through architectural transitions. He positions Tirias as a custom research and advisory shop that tracks everything from IP blocks and SoCs to full data center and edge deployments across AI, HPC and quantum-ready systems. https://tiriasresearch.com/

—
HDMI® Technology is the foundation for the worldwide ecosystem of HDMI-connected devices; integrated with displays, set-top boxes, laptops, audio video receivers and other product types. Because of this global usage, manufacturers, resellers, integrators and consumers must be assured that their HDMI® products work seamlessly together and deliver the best possible performance by sourcing products from licensed HDMI Adopters or authorized resellers. For HDMI Cables, consumers can look for the official HDMI® Cable Certification Labels on packaging. Innovation continues with the latest HDMI 2.2 Specification that supports higher 96Gbps bandwidth and next-gen HDMI Fixed Rate Link technology to provide optimal audio and video for a wide range of device applications. Higher resolutions and refresh rates are supported, including up to 12K@120 and 16K@60. Additionally, more high-quality options are supported, including uncompressed full chroma formats such as 8K@60/4:4:4 and 4K@240/4:4:4 at 10-bit and 12-bit color.
—

He characterizes Supercomputing 2025 #SC25 as a turning point where classical high-performance computing, large-scale AI and early quantum computers are converging into a single conversation. Using IBM’s “quantum chandelier” as a visual anchor, McGregor explains how competing qubit technologies could eventually yield a single quantum processor capable of outperforming today’s Top500 supercomputers on targeted workloads. That leap unlocks scenarios like real-time global weather modeling, accelerated materials discovery and fast genomic analysis that are impractical even on the largest current clusters.

Quantum also reshapes security. McGregor emphasizes the “harvest now, decrypt later” threat model, where attackers capture encrypted traffic today and wait for future quantum systems to break current public-key schemes. He argues that organizations should already be migrating to NIST-approved post-quantum cryptography, designing protocols, key management and hardware roots of trust so that sensitive data remains protected even when practical quantum accelerators become widely available.

On the classical compute side, he discusses Nvidia’s Arm-based Grace server CPUs and upcoming Vera/Blackwell-era platforms as tightly coupled companions to GPU accelerators for AI-centric nodes, while also noting competing heterogeneous designs from AMD and others. He links these product trends to a wave of new exascale projects, including future US Department of Energy systems and additional AMD-powered exascale machines in Europe, as vendors push toward AI-optimized “AI factory” data centers rather than traditional monolithic supercomputers.

McGregor closes by walking through the rest of the SC25 show floor to illustrate how every subsystem is scaling together: Micron’s 245 TB SSDs for dense storage, high-bandwidth networking from companies like Broadcom and Cisco, and visually striking liquid-cooling deployments that make ultra-dense AI racks thermally feasible. Filmed at Supercomputing 2025 in St. Louis, the discussion frames the modern data center as a single engineered system where memory, interconnects, accelerators, cooling and power delivery all co-evolve to meet the demands of HPC, AI and, increasingly, quantum workloads.

I’m publishing about 90+ videos from Embedded World North America 2025, I upload about 4 videos per day at 5AM/11AM/5PM/11PM CET/EST. Join https://www.youtube.com/charbax/join for Early Access to all 90 videos (once they’re all queued in next few days) Check out all my Embedded World North America videos in my Embedded World playlist here: https://www.youtube.com/playlist?list=PL7xXqJFxvYvjgUpdNMBkGzEWU6YVxR8Ga

This video was filmed using the DJI Pocket 3 ($669 at https://amzn.to/4aMpKIC using the dual wireless DJI Mic 2 microphones with the DJI lapel microphone https://amzn.to/3XIj3l8 ), watch all my DJI Pocket 3 videos here https://www.youtube.com/playlist?list=PL7xXqJFxvYvhDlWIAxm_pR9dp7ArSkhKK

Click the “Super Thanks” button below the video to send a highlighted comment under the video! Brands I film are welcome to support my work in this way 😁

Check out my video with Daylight Computer about their revolutionary Sunlight Readable Transflective LCD Display for Healthy Learning: https://www.youtube.com/watch?v=U98RuxkFDYY

source https://www.youtube.com/watch?v=muBlRN2zL0g