MemryX MX3 edge AI accelerator delivers up to 5 TOPS, is offered in die, package, and M.2 and mPCIe modules

MemryX MX3 EVB

Jean-Luc noted the MemryX MX3 edge AI accelerator module while covering the DeGirum ORCA M.2 and USB Edge AI accelerators last month, so today, we’ll have a look at this AI chip and corresponding modules that run computer vision neural networks using common frameworks such as TensorFlow, TensorFlow Lite, ONNX, PyTorch, and Keras. MemryX MX3 Specifications MemryX hasn’t disclosed much performance stats about this chip. All we know is it offers more than 5 TFLOPs. The listed specifications include: Bfloat16 activations Batch = 1 Weights: 4, 8, and 16-bit ~10M parameters stored on-die Host interfaces – PCIe Gen 3 I/O and/or USB 2.0/3.x Power consumption – ~1.0W 1-click compilation for the MX-SDK when mapping neural networks that have multiple layers Under the hood, the MX3 features MemryX Compute Engines (MCE) which are tightly coupled with at-memory computing. This design creates a native, proprietary dataflow architecture that utilizes up to 70% […]

NXP i.MX 95 SMARC 2.1 system-on-modules – ADLINK LEC-IMX95 and iWave iW-RainboW-G61M

SMARC 2.1 development board NXP i.MX95

Several companies have unveiled SMARC 2.1 compliant system-on-modules powered by the NXP i.MX 95 AI SoC, and today we’ll look at the ADLINK LEC-IMX95 and iWave Systems iW-RainboW-G61M and related development/evaluation kits. The NXP i.MX 95 SoC was first unveiled at CES 2023 with up to six Cortex-A55 application cores, a Cortex-M33 real-time core, and a low-power Cortex-M7 core, as well as an eIQ Neutron NPU for machine learning applications. Since then a few companies have unveiled evaluation kits and system-on-modules such as the Toradex Titan evaluation kit or the Variscite DART-MX95 SoM, but none of those were compliant with a SoM standard, but at least two SMARC 2.1 system-on-modules equipped with the NXP i.MX 95 processor have been introduced. ADLINK LEC-IMX95 Specifications: SoC – NXP i.MX 95 CPU Up to 6x Arm Cortex-A55 application cores clocked at 2.0 GHz with 32K I-cache and D-cache, 64KB L2 cache, and 512KB […]

GEEKOM A8 (AMD Ryzen 9 8945HS) AI mini PC review – Part 1: Specs, unboxing, teardown, and first boot

GEEKOM A8 review Windows 11 Pro

GEEKOM A8 is an AI mini PC based on the powerful AMD Ryzen 9 8945HS (or Ryzen 7 8845HS) AI processor with AMD Radeon 780M Graphics, up to 64GB DDR5 memory, up to 2TB M.2 NVMe SSD support for up to four display up to 8K resolution, and comes preloaded with Windows 11 Pro operating system. The mini PC is equipped with two HDMI 2.1 ports, two USB-C ports with DisplayPort Alt mode, 4x USB 3.2 Type-A ports, 2.5GbE, a WiFi 6E and Bluetooth 5.3 module, and a stereo headset jack. GEEKOM sent us a sample of the A8 Mini PC with an AMD Ryzen 9 8945HS 8-core/16-thread processor, 32GB DDR5, and a 2TB M.2 NVMe SSD with Windows 11 Pro for review this time. We’ll start by listing some specifications, doing an unboxing, going through a teardown, and booting Windows 11. In the second and third parts of the […]

New NXP i.MX 93-based system-on-modules launched by MYiR, Variscite, and Compulab

MYIR MYD-LMX9X development board

We have covered announcements about early NXP i.MX 93-based system-on-modules such as the ADLINK OSM-IMX93 and Ka-Ro Electronics’ QS93, as well as products integrating the higher-end NXP i.MX 95 processor such as the Toradex Titan Evaluation kit. Three additional NXP i.MX 93 SoMs from Variscite, Dart, and Compulab are now available. Targeted at industrial, IoT, and automotive applications, the NXP i.MX 93 features a 64-bit dual-core Arm Cortex-A55 application processor running at up to 1.7GHz and a Cortex-M33 co-processor running at up to 250MHz. It integrates an Arm Ethos-U65 microNPU, providing up to 0.5TOPS of computing power, and supports EdgeLock secure enclave, NXP’s hardware-based security subsystem. The heterogeneous multicore processing architecture allows the device to run Linux on the main core and a real-time operating system on the Cortex-M33 core. The processor is designed for cost-effective and energy-efficient machine learning applications. It supports LVDS, MIPI-DS, and parallel RGB display protocols […]

Banana Pi BPI-F3 SBC features SpacemIT K1 octa-core RISC-V AI SoC

Banana Pi BPI-F3 SBC

Banana Pi BPI-F3 single board computer (SBC) is powered by the same SpacemiIT K1 octa-core 64-bit RISC-V SoC with 2TOP AI accelerator found in the upcoming Muse Book RISC-V laptop. The board comes with up to 4GB RAM and 16GB eMMC flash, supports NVMe or SATA storage via its M.2 socket, is equipped with HDMI and MIPI DSI display interfaces, two MPI CSI camera interfaces, two gigabit Ethernet ports, a WiFi 5 and Bluetooth 4.2 module, and can also take a PCIe module for 4G LTE cellular connectivity. Other features include four USB 3.0 Type-C ports, a microSD card slot, a 26-pin GPIO header, and optional support for PoE. Banana Pi BPI-F3 specifications: SoC – SpacemiT K1 CPU – 8-core X60 RISC-V processor with single-core performance equivalent to about 1.3x the performance of an Arm Cortex-A55 GPU – Imagination IMG BXE-2-32 with support for OpenCL 3.0, OpenGL ES3.2, Vulkan 1.2 […]

XGO-Rider is a 2-wheel self-balancing robot with an ESP32 controller plus either a Raspberry Pi CM4 or BBC Micro:bit (Crowdfunding)


XGO-Rider is a two-wheel self-balancing robot with an ESP32 controller for motor and servo control, USB-C charging, etc… and a choice between a Raspberry Pi CM4 module or a BBC Micro:bit board for display, audio, and camera (CM4-only). It’s not the first robot from Luwu Intelligence, since the company launched the XGO-Mini robot dog in 2021, followed by the XGO 2 Raspberry Pi CM4-powered desktop robotic dog with an arm which we reviewed last year. The new XGO-Rider builds on these earlier models but in a different form factor moving from four-legged robots to a 2-wheel self-balancing robot design with many of the same features including AI vision running on the Raspberry Pi CM4. XGO-Rider specifications: Host controller (one or the other) Raspberry Pi CM4 with 2GB RAM + ESP32 for main control, USB-C charging port, DIP switch BBC Micro:bit V2 + ESP32 for main control, USB-C charging port, DIP […]

Sipeed MaixCAM is a RISC-V AI camera devkit with up to 5MP camera, 2.3-inch color touchscreen display, GPIOs

Sipeed MaixCAM

Sipeed MaixCAM is an AI camera based on SOPHGO SG2002 RISC-V (and Arm, and 8051) SoC with a 1 TOPS NPU that takes up to 5MP camera modules and comes with a 2.3-inch color touchscreen display. The development kit also comes with WiFi 6 and BLE 5.4 connectivity, optional Ethernet, audio input and output ports, a USB Type-C port, and two 14-pin GPIO headers for expansion that makes it suitable for a range of computer vision, Smart audio, and AIoT applications. Sipeed MaixCAM specifications: SoC – SOPHGO SG2002 CPU 1 GHz RISC-V C906 processor or Arm Cortex-A53 core (selectable at boot) running Linux 700 MHz RISC-V C906 core running an RTOS 25 to 300 MHz low-power 8051 processor NPU – 1 TOPS @ INT8 with support for models such as Mobilenetv2, YOLOv5, YOLOv8, etc… Video Codec – H.264, H.265, MJPEG hardware encoding and decoding up to 2K @ 30fps Memory […]

BitNetMCU project enables Machine Learning on CH32V003 RISC-V MCU

Neural networks on the CH32V003

Neural networks and other machine learning processes are often associated with powerful processors and GPUs. However, as we’ve seen on the page, AI is also moving to the very edge, and the BitNetMCU open-source project further showcases that it is possible to run low-bit quantized neural networks on low-end RISC-V microcontrollers such as the inexpensive CH32V003. As a reminder, the CH32V003 is based on the QingKe 32-bit RISC-V2A processor, which supports two levels of interrupt nesting. It is a compact, low-power, general-purpose 48MHz microcontroller that has 2KB SRAM with 16KB flash. The chip comes in a TSSOP20, QFN20, SOP16, or SOP8 package. To run machine learning on the CH32V003 microcontroller, the BitNetMCU project does Quantization Aware Training (QAT) and fine-tunes the inference code and model structure, which makes it possible to surpass 99% test accuracy on a 16×16 MNIST dataset without using any multiplication instructions. This performance is impressive, considering […]

Exit mobile version
EmbeddedTS embedded systems design