Hardkernel Launches the ODROID-M2, Promises Three Times the Performance of the Original ODROID-M1
Offering a claimed threefold increase in CPU and NPU and fivefold increase in GPU performance, the ODROID-M2 looks like a worthy successor.
Single-board computer specialist Hardkernel has unveiled a new entry in its ODROID family, designed as a more powerful successor to the ODROID-M1: the sensibly-named ODROID-M2, powered by a Rockchip RK3588S2 system-on-chip.
"We launched the ODROID-M1 about two years ago, and the ODROID-M1S about one year ago," Hardkernel explains by way of background to the launch. "Both models have been successfully supplied and continue to be adopted as core components in embedded systems by our B2B [Business-to-Business] customers. We received feedback that the performance and input/output port configuration of the ODROID-M1/M1S were sufficient for most embedded systems. However some customers still wanted higher-end models equipped with high-performance processors. To meet the demand for higher performance computing power required by important industrial embedded system builders, we are launching the new ODROID-M2 SBC today."
The launch comes around nine months after the company celebrated 15 years in the business with a shrunken version of the original ODROID-M1, dubbed the ODROID-M1S. Where that boasted roughly same specifications as its predecessor, with a quad-core Arm Cortex-A55 running at a slightly slower 1.8GHz and with the loss of a few ports, the ODROID-M2 is a fully new device — around three times the speed of the original, workload dependent.
That performance gain comes primary from a move from the Rockchip RK3568B2 system-on-chip to the RK3588S2, which has four Arm Cortex-A55 cores running at 1.8GHz alongside four higher-performance Cortex-A76 cores running at 2.3GHz — give or take a hundred megahertz, Hardkernel says. The chip also includes an Arm Mali-G610 MP4 graphics processor running at 1GHz and a RKNN neural network coprocessor for accelerating on-device machine learning and artificial intelligence (ML and AI) workloads, delivering a claimed six tera-operations per second (TOPS) of compute at INT8 precision.
There's a choice of 8GB or 16GB or LPDDR5 memory, model depending, and 64GB of eMMC — soldered to the board, rather than fitted as a user-swappable module. A microSD Card slot offers UHS-I SDR104 performance, and there's an M.2 slot for a Non-Volatile Memory Express (NVMe) SSD connected over a single PCI Express Gen. 2.1 lane. For display connectivity there's an HDMI 2.0 port supporting 4k60 output, a MIPI Display Serial Interface (DSI) port — which uses a 30-pin connector, incompatible with the 31-pin connector on the ODROID-M1 — and DisplayPort over a USB Type-C connector.
Speaking of USB, there's one each of USB 2.0 and USB 3.0 hosts ports along with the aforementioned USB 3.0/DisplayPort Type-C connector, alongside a Raspberry Pi-style 40-pin general-purpose input/output (GPIO) header and a second 14-pin GPIO header — plus a dedicated connector for a UART debug console.
The company claims that the board offers around triple the performance of the original ODROID-M1 in multi-threaded processor-centric workloads, over five times the performance for GPU workloads, and three times the performance for workloads on the neural coprocessor. Power draw has been measured at 1W idle and 7.5W under CPU load, with its DC jack accepting anything from 7.5V to 15.5V.
The ODROID-M2 is now available for sale on the Hardkernel store, priced at $115 for the 8GB version and $145 for the 16GB version; a 12V 2A power supply is available to order at the same time for $5.50 extra.