In this interview from Supercomputing SC25 in St. Louis, SiMa.ai and Advantech frame “physical AI” as an edge-first compute problem: many camera and sensor streams, strict latency, and a power budget that looks nothing like a data center. The demo centers on an Advantech edge server populated with five SiMa.ai MLSoC PCIe accelerator cards, pushing real-time analytics across 80 video channels in a single box, with each card handling up to 16 Full HD 30 fps streams end-to-end. https://sima.ai/
—
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.
—
A key point is that the pipeline is not just running neural networks after someone else decodes video. The MLSoC approach pulls decode, pre/post processing, CNN inference, and re-encode onto the accelerator so an x86 host doesn’t become the choke point when channel counts climb. SiMa.ai describes this as heterogeneous compute on one SoC: Arm cores for control/HMI, a vector and computer-vision engine for classic algorithms, plus a purpose-built ML accelerator that can execute both CNN and transformer-style models on one chip.
On the software side, the pitch is a single environment rather than a pile of glue code. Palette is positioned as the suite to develop, compile, deploy, and maintain edge AI apps across PCIe, system-on-module, and chip-down designs, while Edgematic adds a more visual, low-code way to assemble pipelines and validate IO and model behavior. For GenAI, SiMa.ai has been promoting LLiMa inside Palette to automate quantization, memory-aware compilation, and packaging for on-device LLM/VLM execution in one automated flow.
The scaling story is intentionally practical: add cards to add streams, or split mixed workloads across cards (for example segmentation on one card and conversational AI on another) with scheduling handled by the stack. SiMa.ai’s half-height/half-length PCIe board briefs describe low-power operation (often around 10–15 W on typical workloads) and PCIe Gen4 up to x8, while Advantech’s SC25 server material emphasizes large DDR5 capacity and multiple expansion slots to host these accelerators densely without turning the chassis into a GPU rack.
Overall, the conversation is less about peak TOPS and more about throughput-per-watt for “always on” deployments where video IO, memory movement, and deployment friction decide whether an AI feature ships. It’s also a reminder that edge GenAI isn’t only about tokens-per-second—it’s about fitting transformers into real-time, multi-stream perception stacks that need deterministic latency, manageable updates, and predictable power draw, shown here in a partner demo at SC25 there.
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
