ipXchange’s Post

View organization page for ipXchange, graphic

7,601 followers

1mo

How to run GPU-level edge AI on a Cortex-M MCU – NVIDIA and Edge Impulse’s latest tools At Hardware Pioneers Max 2024, ipXchange chatted with Edge Impulse’s Amir Sherman about NVIDIA TAO and NVIDIA Omniverse, two AI tools that have enabled Edge Impulse’s customers to run GPU-level edge AI workloads on low-cost, low-power general-purpose microcontrollers. So how does a Renesas Electronics Cortex-M85 MCU measure against an NVIDIA Jetson Orin GPU for image classification, and how does Edge Impulse make it possible to run the same AI task on both devices? Keep designing! #embedded #edgeAI #AI #artificialintelligence #AIvision #AIcamera #electronics #electronicsengineering #MCU #CortexM #microcontroller #CortexM85 #CortexM55 #CortexM33 #IoT #AIoT #industrialelectronics #IndustrialAI #GPU #Jetson

Transcript

Hey, Simon with IP Exchange, a hardware Pioneers Max 2024. I'm here talking to with Amir at the Edge Impulse booth. So we're gonna do a bit of a run through of edge Impulse in action. But first, I mean, would you like to introduce yourself and what you do at edging pots? Yes, thank you very much. Thank you for having me. My name is Amira Sherman. I'm senior director global business development for semiconductors and eco partners and I'm happy to be here at how the Pioneer in London. Cool. So if we can start first, what's the hardware we're using? And what's the use case? OK, so specifically today I would like to highlight something that we have been working in the last one year and been announced in the last NVIDIA GTC. So think about it, what you see here, this is an Advantech based production ready camera is inside, you have an Invidia Jetson Orange GPU. OK, so literally customers are able to buy this, use it for production running, you know, huge amount of AI modeling in production unit. Magic that we are going to talk today in my lecture and I will show you a little bit is that you are able to one AI model the same a model that run these massive GPU on on the microcontroller. OK, microcontroller written without a accelerating it's amazing. Wow, this is the platform. So out of interest, what microcontroller is that one? So in this specific case, you know, we have a general purpose microcontroller Army base, you know ITOS accelerated M33M55 or even the latest M85. We can without accelerator, but these are, you know, low power $510 microcontrollers running massive, you know, gigantic AI model that was running on NVIDIA GPUs. Cool. So I mean, obviously that solves quite a big problem in terms of power consumption. Exactly how do you get started porting such a model to a microcontroller? How do you start training it? OK, excellent. So before talking about the training, we connected with NVIDIA. OK, NVIDIA created the special toolkit. All NVIDIA Tau toolkit OK which platform and the idea is that Invidia for sure is it dominated in the a market understand that their Jetson will be never you know low power there will be never $10. This is not their market yeah, yeah. But in the end they want to play in this market how they're doing it. They created the software platform calling video OK and this Tau platform is a way to simplify, quantize shrink big AI. OK, that will be able to run on small microcontrollers, microprocessors, AI accelerators in the market. What is for NVIDIA in the end, you need to train the models. The training will be all the time in the cloud, utilizing the GPU's in the cloud. It's a win win for everybody. Yeah, cool. So yeah, how what does it look like when you get started? OK, so in the end I will not go into the details, you know, platform unable to collect data or bring your data and then create. AI modeling, but here's specifically what you see. We already enabled the NVIDIA Tau toolkit part of our platform. So customers are able to choose which type of Invidia town model they want to deploy Retina net, yellow 3, yellow four and tower SSD. Now specifically, let's look in in this side. What you see here is an example of a forklift where there is an computer vision to recognize pallets in industrial application. This is will customer now. This demo was connected with the real Jetson Orinando that was hanging in the in in the in the in the factory, OK. And you can see from the first second he recognized all the pallets, you know, running a lot of frame per seconds. Everything is perfect. This is the perfect world. The only thing is that this is a $500 product, 20 watts very, you know, consuming, you know, very big, the same model that was running. Here is running in this case on a Renaissance latest ARM Cortex M85. Now let's say in the beginning here we recognize here he didn't recognize the palette and now there is a one pallet here and there is another pallet here. And then when it's getting closer and closer and closer, he recognize all the pallets. And the message here is that what good enough can be good enough? Meaning this is a massive GPU, this is a microcontroller if you add more data to it. You can get more accuracy and then you then you can use that how you add more data. We also connected with NVIDIA Omniverse. NVIDIA Omniverse is a synthetic data platform, meaning more data, better AI model, synthetic data that connected with real data will help this model to be better. And then in the end, customers can use this into production. They start with that. We'll finish with this and this is the magic. Cool. So I guess if I was to summarize that, would you say? Accurate to say ohh one you're enabling something that used to be done on a G on an Invidia GPU to be done on a microcontroller act and to with this Omniverse platform, you don't need to have all the real world data you can have synthetic data data exactly improve the the model so that it works good enough for the application still has that low power consumption. Exactly this is it and this is the magic and we are now walking to add more deployment to general. That this microcontrollers microprocessor we knew and Linux machines and this is the magic of wattage input is doing and present later on today. Cool. So can you just give me a little bit of information of what you will be talking about in the talk? Yeah. So specifically on my talk is you know how to drive in NVIDIA tell models and any GPU, any from GPU's to any microcontrollers and processors. And I will literally highlight this demonstration to the audience talking a little bit more about what is coming. So we are going to add more models and more new announcements about you. Microcontroller companies that will be part of the platform. Cool. So if you really want to shrink the power consumption of your age, I then send edge impulse all the way and you can do that on a microcontroller now. So yeah, check this out and we'll have the full right up on the Exchange website.

To view or add a comment, sign in

More Relevant Posts

Edge Impulse

41,352 followers
1mo
Report this post
Bring the power of NVIDIA AI TAO models from GPUs to microcontrollers with Edge Impulse. This interview with Amir Sherman is a great overview of how we can take computer vision models and shrink them down to fit on MCUs, enabling low-power inference on inexpensive devices. #EdgeAI #MachineLearning #Nvidia #AI

ipXchange

7,601 followers
1mo

How to run GPU-level edge AI on a Cortex-M MCU – NVIDIA and Edge Impulse’s latest tools At Hardware Pioneers Max 2024, ipXchange chatted with Edge Impulse’s Amir Sherman about NVIDIA TAO and NVIDIA Omniverse, two AI tools that have enabled Edge Impulse’s customers to run GPU-level edge AI workloads on low-cost, low-power general-purpose microcontrollers. So how does a Renesas Electronics Cortex-M85 MCU measure against an NVIDIA Jetson Orin GPU for image classification, and how does Edge Impulse make it possible to run the same AI task on both devices? Keep designing! #embedded #edgeAI #AI #artificialintelligence #AIvision #AIcamera #electronics #electronicsengineering #MCU #CortexM #microcontroller #CortexM85 #CortexM55 #CortexM33 #IoT #AIoT #industrialelectronics #IndustrialAI #GPU #Jetson
Like Comment
To view or add a comment, sign in
OKdo

9,324 followers
1mo
Report this post
🚀 Ready to Transform Your #AI Projects? Powered by the #NVIDIA Jetson Nano Module, the OKdo Nano C100 #DeveloperKit is designed to deliver exceptional compute performance for your #ArtificialIntelligence and #MachineLearning workloads. Here are the top 5 reasons why this developer kit should be your next investment: 🌟 Unparalleled Compute Power - Harness the power of the 128-core NVIDIA® Maxwell GPU and Quad-core ARM A57 CPU to run advanced AI frameworks with ease. 🔌 Diverse Connectivity Options - Connect a wide range of sensors and peripherals with extensive I/Os, including GPIO, CSI, HDMI, and more. 🔋 Energy-Efficient Operation - Operates on as little as 5 watts, perfect for power-sensitive applications without compromising performance. 💻 Robust Software Ecosystem - Fully supported by NVIDIA® JetPack, featuring CUDA, cuDNN, and TensorRT, enabling streamlined AI development and deployment. ⚡ Versatile Power Solutions - Choose between micro-USB or a DC power adapter to meet your project requirements. Whether you're a pro engineer, student, or developer, the OKdo Nano C100 Developer Kit is your gateway to cutting-edge AI innovation. Get ready to push the boundaries of what's possible! #JetsonNano #IoT #TechInnovation #LetsOKdo Buy Now: https://weare.rs/4d3UzJq
Like Comment
To view or add a comment, sign in
Chloe Jian Ma
1mo
Report this post
AI will go beyond perception and reasoning towards taking actions to further assist human beings with dirty, dull and dangerous tasks. That is why #robotics is all the rage now. Get your hands dirty on #robotics and #automation with less than $100 from Seeed Studio #onArm. #edgeAI #AI
Seeed Studio

40,614 followers
1mo Edited

Looking for an open-source, budget-friendly SBC that's perfect for #robotics, building #automation, or intelligent #HMI for vision tasks? You won't want to miss BeagleY®-AI! It is an advanced SBC with an industry-standard form factor developed by BeagleBoard.org Foundation. Powered by the Texas Instruments #AM67A ARM-based Vision Processor, it's now fully stocked in our US warehouse, priced at just $70 per unit. And here's the best part: when you buy 2 units, you'll enjoy free shipping! 👉 https://bit.ly/456mvcR 💡 Why Choose BeagleY®-AI? ✅ 4 TOPS AI Performance for advanced deep learning ✅ 4GB LPDDR4, Micro SD card, and Board identifier EEPROM ✅ Quad-core 64-bit Arm® Cortex®-A53 CPU at 1.4GHz. ✅ Flexible I/O including PCIe, USB3, and Ethernet with PoE support ✅ 40-pin expansion header for endless prototyping We extend our special thanks to Jason Kridner and Christine Long, who have contributed their expertise to bring you these amazing products once again. #edgecomputing #edgedevice #edgeiot #embedded #AI #AIoT #iiot #iot #objectdetection #imageclassification #semanticsegmentation
Like Comment
To view or add a comment, sign in
PT Wellracom Industri Komputindo

342 followers
7mo
Report this post
Axiomtek – a world-renowned leader relentlessly devoted to the research, development, and manufacture of series of innovative and reliable industrial computer products of high efficiency – is proud to unveil the AIE110-ONX, its new fanless edge AI developer kit that provides up to 100 TOPS of accelerated AI computing performance. The AIE110-ONX is powered by the NVIDIA® Jetson Orin™ NX system-on-module which has a 6-core/8-core NVIDIA ARM® Cortex A78AE v8.2 64-bit CPU and a 1024-core NVIDIA Ampere GPU with 32 Tensor Cores. This small yet powerful edge AI developer kit, tailored for quick development with no minimum order, is well-suited for makers, learners, and system integrators seeking to effortlessly enhance AI innovation. #wellracom #IOT #automation #CANopen #PLC #SCADA #IIOT #Modbus #OPCUA #automation #data #network #transformation #water #communication #communications #communications #utilities #ai #embeddedsystems #embeddedsystems #people #iot #manufacturing #artificialintelligence #agriculture #analytics #retail #productivity #power #surveillance #computervision #transportation #video #smartcities #mikrotik #routeros
Like Comment
To view or add a comment, sign in
Dharun Kumar

Autonomous systems🤖| MSc MPSYS at Chalmers University| B. Tech ECE @ ASE
4mo
Report this post
Cheaper options for boards for AI computing. 1. Arduino Nano BLE Sense 33 - Ideal for implementing AI-powered IoT applications with built-in environmental sensors and Bluetooth connectivity. https://encr.pw/CDNsr 2. Raspberry Pi Pico - Suitable for AI projects requiring low-cost microcontroller solutions with GPIO flexibility and MicroPython support. https://l1nq.com/5BWgM 3. ESP32 - Well-suited for AI applications involving wireless communication, sensor integration, and edge computing capabilities. https://l1nq.com/8HZfD 4. STM32F7 - Capable of handling complex AI algorithms and real-time processing tasks in embedded systems with its high-performance ARM Cortex-M7 core. https://acesse.dev/bXRLR 5. Teensy 4.0 - Enables high-speed AI inference and sensor integration in compact and power-efficient designs, suitable for robotics and wearable devices. https://encr.pw/LzvmP 6. Nvidia Jetson Nano - Empowers AI projects with advanced machine learning capabilities and GPU acceleration for tasks like computer vision and robotics. https://l1nq.com/1UCEs Want to know more? Let's connect🫂 and learn together📕 #learning #resources #ai
4 Comments
Like Comment
To view or add a comment, sign in
Seeed Studio

40,614 followers
1mo Edited
Report this post
Looking for an open-source, budget-friendly SBC that's perfect for #robotics, building #automation, or intelligent #HMI for vision tasks? You won't want to miss BeagleY®-AI! It is an advanced SBC with an industry-standard form factor developed by BeagleBoard.org Foundation. Powered by the Texas Instruments #AM67A ARM-based Vision Processor, it's now fully stocked in our US warehouse, priced at just $70 per unit. And here's the best part: when you buy 2 units, you'll enjoy free shipping! 👉 https://bit.ly/456mvcR 💡 Why Choose BeagleY®-AI? ✅ 4 TOPS AI Performance for advanced deep learning ✅ 4GB LPDDR4, Micro SD card, and Board identifier EEPROM ✅ Quad-core 64-bit Arm® Cortex®-A53 CPU at 1.4GHz. ✅ Flexible I/O including PCIe, USB3, and Ethernet with PoE support ✅ 40-pin expansion header for endless prototyping We extend our special thanks to Jason Kridner and Christine Long, who have contributed their expertise to bring you these amazing products once again. #edgecomputing #edgedevice #edgeiot #embedded #AI #AIoT #iiot #iot #objectdetection #imageclassification #semanticsegmentation
2 Comments
Like Comment
To view or add a comment, sign in
Sahil Raut

Robotics Engineer | IEEE RAS Member | BEng(Hons) Robotics
6mo Edited
Report this post
🚀 𝗜𝗻𝘁𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝗼𝗻 𝘁𝗼 𝘁𝗵𝗲 OpenMV LLC 𝗛𝟳 𝗣𝗹𝘂𝘀 𝗖𝗮𝗺: 🚀 🔥 𝗤𝘂𝗶𝗰𝗸 𝗛𝗶𝗴𝗵𝗹𝗶𝗴𝗵𝘁𝘀: 𝗣𝗼𝘄𝗲𝗿𝗳𝘂𝗹 𝗖𝗼𝗿𝗲: STM32H743II ARM Cortex M7 @ 480 MHz. 𝗠𝗮𝘀𝘀𝗶𝘃𝗲 𝗠𝗲𝗺𝗼𝗿𝘆: 32MB SDRAM, 32MB Flash + Extras. 𝗘𝗮𝘀𝘆 𝗣𝗿𝗼𝗴𝗿𝗮𝗺𝗺𝗶𝗻𝗴: High-level Python scripts? Yes, please! 𝗖𝗼𝗻𝗻𝗲𝗰𝘁𝗶𝘃𝗶𝘁𝘆 𝗚𝗮𝗹𝗼𝗿𝗲: USB, μSD, SPI, I2C, CAN, Serial, ADC/DAC. 𝗖𝗮𝗺𝗲𝗿𝗮 𝗙𝗹𝗲𝘅𝗶𝗯𝗶𝗹𝗶𝘁𝘆: Comes with a 5MP OV5640 sensor. Swap for global shutter or thermal vision! 𝗦𝗲𝗿𝘃𝗼 & 𝗟𝗘𝗗 𝗥𝗲𝗮𝗱𝘆: Drive servos, control an RGB LED, and IR LEDs. 𝗕𝗮𝘁𝘁𝗲𝗿𝘆 𝗣𝗼𝘄𝗲𝗿𝗲𝗱: Hook up a 3.7V LiPo for mobile projects. 📚 For more info About the board: https://lnkd.in/e3iQBBqH Other references: https://lnkd.in/e4PMqPaR Do let me know your thoughts on this board if you would invest in this for your projects and how? #OpenMV #IoT #EmbeddedSystems #CircuitPython #STM32H743II #AI #Innovation #robotics

7 Comments
Like Comment
To view or add a comment, sign in
F&S Elektronik Systeme GmbH

627 followers
7mo
Report this post
👉 Coming soon: PicoCore SoM board family series based on NXP i.MX95: Powering the Future of Innovation together with us! 👈 Enhanced Performance: With advanced Arm® Cortex®-A cores and powerful graphics capabilities, the PicoCoreMX95 series offers exceptional performance for demanding applications, ensuring seamless user experiences. AI and Machine Learning: Leveraging dedicated AI accelerators, these modules enable cutting-edge AI and machine learning applications, propelling innovation in various industries. Robust Security Features: Security is paramount in today's interconnected world. The PicoCoreMX95 module series integrates robust security features, safeguarding sensitive data and ensuring a trustworthy environment. Versatility and Connectivity: Designed for versatility, these boards support a wide range of interfaces and connectivity options, facilitating seamless integration into diverse systems. NXP Semiconductors NXP Partner Channel #NXP #iMX95 #Innovation #EmbeddedProcessing #Technology #AI #IoT #Semiconductors #Madeingermany #SoM #ublox #microsoft #windowsiot #arm #azure #edgecomputing #edgedevice #stuttgart #SGet #NXP #eIQAI #ArtificialIntelligence #AI #MachineLearning #EdgeAI #EmbeddedAI #AIApplications #DeepLearning #NXPInnovation #IoT #AIPlatforms #EdgeComputing #EmbeddedSystems #eIQSoftware #EmbeddedApplications #CPUs #Semiconductors #EdgeComputing #EmbeddedSystems #NXPProcessors #NXPPartner #NXPGoldPartner #arm #armcpu #PicoCore #efus #armStone #SolderCore #OSM #EPD #Madeingermany
Like Comment
To view or add a comment, sign in
Yvonne Yu

ISO9001 certificated Distributor of electronic components 11+ years, focus on #TI #MICRON #Microchip #NXP #STM, #Aanlog Devices, and electronic manufacturing's shortage & hard find part service.
1mo
Report this post
#Microchip Launches PIC64 Portfolio for Embedded and Space Apps Microchip Technology introduced its PIC64 portfolio to address the growing need for power efficiency, hardware-level security, and high reliability in real-time, compute-intensive applications like smart embedded vision and machine learning (ML). The PIC64 family aims to provide a single-vendor solution for microprocessors (MPUs), catering to markets requiring both real-time and application-class processing. The first release in this new product line, the PIC64GX MPUs, supports intelligent edge designs across various sectors, including industrial, automotive, communications, IoT, aerospace and defense. Moreover, Microchip announced the PIC64 High-Performance #Spaceflight Computing (PIC64-HPSC) MPUs to meet the growing computational needs of the expanding space market. Experts project the space sector to grow from $344 billion in 2023 to $755 billion by 2035. This evolution follows two decades of advancements since trailblazing processor-powered missions like Deep Impact and Mars Curiosity rover. With the introduction of its #PIC64 portfolio, Microchip has become an embedded solutions provider actively developing a full spectrum of 8-, 16-, 32- and 64-bit #microcontrollers (MCUs) and MPUs. In an interview with EE Times, Shakeel Peera, VP of strategy and marketing for Microchip’s #FPGA business unit, and Venki Narayanan, director of marketing for Microchip’s FPGA business unit, discussed the significance of Microchip’s latest portfolio launch concerning the renowned PIC family.
Like Comment
To view or add a comment, sign in
Monica Lee

(ISO9001,DUNS certificated) Leading Top 20 Gold Distributor Electric components 12 years #NXP #ST #ON #TI #ADI #ALTERA #XILINX #Cypress #Microchip #ST,PCB Solution;Shortage and hard find part ect.
1mo
Report this post
#Microchip Launches PIC64 Portfolio for Embedded and Space Apps Microchip Technology introduced its PIC64 portfolio to address the growing need for power efficiency, hardware-level security, and high reliability in real-time, compute-intensive applications like smart embedded vision and machine learning (ML). The PIC64 family aims to provide a single-vendor solution for microprocessors (MPUs), catering to markets requiring both real-time and application-class processing. The first release in this new product line, the PIC64GX MPUs, supports intelligent edge designs across various sectors, including industrial, automotive, communications, IoT, aerospace and defense. Moreover, Microchip announced the PIC64 High-Performance #Spaceflight Computing (PIC64-HPSC) MPUs to meet the growing computational needs of the expanding space market. Experts project the space sector to grow from $344 billion in 2023 to $755 billion by 2035. This evolution follows two decades of advancements since trailblazing processor-powered missions like Deep Impact and Mars Curiosity rover. With the introduction of its #PIC64 portfolio, Microchip has become an embedded solutions provider actively developing a full spectrum of 8-, 16-, 32- and 64-bit #microcontrollers (MCUs) and MPUs. In an interview with EE Times, Shakeel Peera, VP of strategy and marketing for Microchip’s #FPGA business unit, and Venki Narayanan, director of marketing for Microchip’s FPGA business unit, discussed the significance of Microchip’s latest portfolio launch concerning the renowned PIC family. #Capacitors #ICs #EmbeddedProcessors #Divider #Multiplier #WidebandSynthesizer #OutputRegulator #Microcontroller #CommunicationICs #Isolators #DisplayController #EectronicComponentsProcurement #BuyElectronicParts #ic #chip　#semiconductor #supplier #supplying #distributor #electroniccomponentsdistributor #spot #ti #micron #tdk #nxp #xilinx #intel #ADI #AMD
Like Comment
To view or add a comment, sign in

ipXchange

7,601 followers

View Profile Follow

More from this author

Explore topics