AMD has announced that it has extended ROCm and PyTorch support to the Radeon RX 7900 XT graphics card. This comes ahead of AMD’s Advancing AI event, which is scheduled for early December.

Last month, AMD announced ROCm 5.7 and PyTorch support for the Radeon RX 7900 XTX and the Radeon PRO W7900 GPUs. This marked the first official support for ROCm/PyTorch on RDNA3 graphics cards. Now, the support has been expanded to include the Radeon RX 7900 XT. AMD has only provided ROCm support for the 7900 level products, with no official communication regarding support for lower-end Radeon RX 7000 series and Radeon Pro 7000 series products.

Source: Phoronix.

CanMV-K230 is a credit card-sized development board for AI and computer vision applications based on the Kendryte K230 dual-core C908 64-bit RISC-V processor with built-in KPU (Knowledge Process Unit) and various interfaces such as MIPI CSI inputs and Ethernet.

The CanMV-K230 development board features the following specifications:

• SoC – Kendryte K230
  • CPU
    • 64-bit RISC-V processor @ 1.6GHz with RISC-V Vector Extension 1.0, FPU
    • 64-bit RISC-V processor @ 800MHz with support for RV64GCB instruction set
  • AI accelerator
    • KPU with support for INT8 and INT16
    • Compatible with TVM, TensorFlow, Pytorch, ONNX
    • Typical network performance: Resnet 50 ≥ 85fps @ INT8; Mobilenet_v2 ≥ 670fps @ INT8; YoloV5S ≥ 38fps @INT8;
  • “DPU” using 3D structured light depth calculation up to 1280x800 @ 30fps
  • VPU – Video encoding/decoding for H.264/H.265/JPEG/MJPEG up to 4Kp40/4Kp20
• System Memory: 512MB LPDDR3
• Storage – QSPI flash, microSD card slot
• Display – HDMI port, MIPI DSI connector up to 1080p60
• Audio – 3.5mm audio jack
• Camera – Up to 5MP camera, up to 3-ch MIPI CSI inputs
• Networking
  • 10/100Mbps Ethernet RJ45 port
  • WiFi 4 and Bluetooth 4.0 via AP6212 wireless module
• USB – 1x USB 2.0 OTG Type-C port port
• Expansion – 40-pin GPIO header with up to 29x GPIOs, 5x PWM, 4x I2C, 2x UART
• Misc – Buttons, RGB LED
• Power Supply – 5V via USB-C port
• Dimension – 85 x 56 mm

The Kendryte K230 processor offers higher efficiency compared to the K510 and K210 SoCs, with Mobilenet V1, Resnet 50, and YoloV5S achieving up to 341 FPS/TOPS. While the documentation for the CanMV-K230 development board is mostly available in Chinese at this time, there are over 30 AI demos available for users to experiment with. Overall, the CanMV-K230 development board provides a compact solution for AI and computer vision applications, with the Kendryte K230 processor delivering improved performance and efficiency.

Source: CNX Software – Embedded Systems News.

SiFive has introduced two new IP blocks, the Intelligence X390 NPU and the Performance P870 RISC-V core, which are designed for SoCs targeting Generative AI and ML applications.

The Performance P870, previously covered by CNX Software, is a high-performance core that supports the RVA23 RISC-V profile specification, Vector 1.0, and Vector Crypto. It features a six-wide, out-of-order 64-bit core and can be configured with up to a 32-core cluster. The P870 offers over 12 SpecINT2k6/GHz and a 50% peak single-thread performance improvement compared to the previous generation SiFive Performance P670. It also includes SiFive features such as 2x 128b VLEN RVV, vector crypto and hypervisor extensions, IOMMU and AIA, non-inclusive L3 cache, and RISC-V WorldGuard security. The P870 is fully compatible with Google’s platform requirements for Android on RISC-V and supports Linux.

The SiFive Intelligence X390 NPU is built on the U7-Series core with a 64-bit RISC-V ISA and an 8-stage dual-issue in-order pipeline. It supports the RISC-V Vector extension and features a 1024-bit VLEN, 512-bit DLEN, single/dual vector ALU, and the Vector Coprocessor Interface eXtension (VCIX) for direct connectivity of external hardware accelerators. The X390 NPU offers a 4x improvement in vector computation compared to the Intelligence X280 NPU, thanks to its single-core configuration, doubled vector length, and dual vector ALUs.

SiFive envisions the P870 and X390 being paired in powerful RISC-V SoCs with up to 32 cores and 8 AI accelerator cores (along with optional custom accelerators) to enable generative AI in data centers. The company promises a full AI solution with optimized libraries, as well as OpenXLA and NN models. Additionally, the Performance P870 is suitable for DPU, storage, networking, 5G applications, and automotive SoCs. In the automotive market, the P870-A RISC-V core can be integrated with the Intelligence X280 AI accelerator and SiFive S7-A cores for ASIL B/D compliance, supporting applications such as standalone ADAS, central compute, and in-vehicle infotainment (IVI).

SiFive also suggests that the P870 and X390 may find their way into consumer applications like edge AI security cameras, mobile devices, wearables, and smart TVs, although no specific details were provided.

For more information, please visit the product page and press release.

Source: CNX Software – Embedded Systems News.

AMD has released ROCm 5.7, the latest version of its GPU compute stack, Phoronix reports. The update includes a new “hipTensor” library, extended support for the ROCgdb debugger with Fortran and OMPD, and optimizations to the rocRAND and MIVisionX libraries. However, there is no official support for new RDNA3 GPUs yet, with speculation that it will be included in ROCm 6.0. The release notes for ROCm 6.0 suggest that it will have fundamental changes and will not be backwards compatible with ROCm 5.x. In terms of supported hardware, only the Radeon VII is officially supported, along with some Radeon Pro and AMD Instinct products. ROCm 5.7 also brings improvements to HIP printf() handling, a beta release of LLVM Address Sanitizer with GPU support, optimizations, and various fixes. More details can be found in the ROCm 5.7 release documentation.

Source: Phoronix.

The Dusun DSGW-380, also known as the Dusun Pi 5, is an industrial AIoT gateway powered by the Rockchip RK3588 octa-core processor. This gateway supports a wide range of connectivity options, making it suitable for various server and network applications.

One of the notable features of the Dusun DSGW-380 is its support for multiple connectivity options, including WiFi 6, Bluetooth LE, 5G, and LoRaWAN. It also comes with dual gigabit Ethernet ports, RS232 and RS485 wired connectivity, and a variety of wireless protocols. With these options, users can easily connect and integrate different devices and systems into their network.

In terms of specifications, the Dusun DSGW-380 offers 8GB of LPDDR4 memory and up to 128GB of eMMC flash storage. It operates in a wide temperature range of -25 to +75°C, making it suitable for industrial environments. The gateway also includes 2 HDMI ports for video output and supports various industry protocols such as BACnet, Profinet, Ethernet/IP, Modbus, and OPC/UA. Dusun DSGW-380 supports DIN Rail or wall mounting, allowing for flexible installation options. The gateway also comes with a variety of ports and interfaces, including USB 3.0 ports, USB 2.0 OTG Type-C port, RS232, RS485, CAN Bus, and I/Os via terminal blocks.

For software compatibility, the Dusun DSGW-380 supports Ubuntu/Debian Linux and ROS. It also features a 6 TOPS AI accelerator that is compatible with popular AI frameworks such as TensorFlow, MXNet, PyTorch, and Caffe. The SDK of the gateway supports the YOLO v3 algorithm, providing users with advanced AI capabilities.

Overall, the Dusun DSGW-380 Rockchip RK3588 AIoT gateway offers a powerful and versatile solution for server and network applications. With its extensive connectivity options, robust specifications, and support for various software frameworks, it is a suitable choice for industrial environments and IoT deployments.

Please note that pricing information for the Dusun DSGW-380 is not publicly available at the time of writing.

Source: CNX Software – Embedded Systems News.