
NXP i.MX 95 Dev Board Adds 10GbE and Dual M.2 Slots
The $300 FRDM-IMX95-PRO pairs a 6-core i.MX 95, 16GB of LPDDR5-6400, an 8 TOPS NPU, and 10GbE networking with two PCIe Gen3 M.2 slots for edge AI.
The First Board Built Around the Bigger i.MX 95 Package
NXP announced the FRDM-IMX95-PRO on August 11, 2026, and the headline is a packaging decision rather than a new chip. Earlier i.MX 95 boards used the 15x15 mm variant of the SoC. This one uses the 19x19 mm package, and those extra millimeters buy pin count — which is how a single-board computer in this class ends up with native 10 Gigabit Ethernet, two M.2 Key-M slots, and a wider memory bus without bolting on external controllers.
- 6x Arm Cortex-A55 at up to 2.0 GHz, plus a Cortex-M7 at 800 MHz and a Cortex-M33 at 333 MHz
- 16GB LPDDR5-6400 and 32GB eMMC 5.1, with microSD 3.0 and 512 Mbit QSPI flash
- 8 eTOPS eIQ Neutron NPU with a Mali-G310 V2 GPU for OpenGL ES 3.2 and OpenCL 3.0
- 10GbE over SlimSAS, two Gigabit RJ45 ports, and two PCIe Gen3 M.2 Key-M slots
- $300, listed as pending stock on NXP's store at announcement
What Silicon Is Inside the FRDM-IMX95-PRO?
The i.MX 95 is a heterogeneous part, and the core mix tells you what it is for. Six Cortex-A55 application cores handle Linux. A Cortex-M7 at 800 MHz handles hard real-time work — motor control, deterministic industrial protocols, anything where a scheduling hiccup is a defect. A Cortex-M33 at 333 MHz sits underneath for low-power and security duties.
On top of that sits the eIQ Neutron NPU rated at 8 eTOPS. That is not a datacenter number and it is not trying to be. It is the range where you run continuous inference on a couple of camera streams — defect detection on a line, object classification on a mobile robot — at a power budget that fits inside a sealed enclosure. Video encode and decode covers 1080p60 and 4Kp30 in H.265 and H.264.
The memory configuration is the quiet upgrade. 16GB of LPDDR5 at 6400 MT/s over a 32-bit bus gives the NPU and the six A55 cores enough bandwidth to work at the same time, which is exactly where earlier boards in this class ran out of room.
Why Do 10GbE and Dual M.2 Matter for Edge AI?
Because the bottleneck at the edge has moved. When a box is aggregating several 4K camera feeds or acting as a gateway for a factory cell, gigabit stops being enough long before the compute does. Native 10GbE over SlimSAS means the board can sit at a network junction rather than behind one.
The two PCIe Gen3 M.2 Key-M slots are the other half of that story. One can take NVMe storage for local buffering; the other can take an AI accelerator if 8 eTOPS turns out to be short for the workload. An optional M.2 Key-E module adds Wi-Fi 6, Bluetooth 5.4, and 802.15.4 for Thread and Zigbee networks.
Physical I/O is generous for a 100 x 90 mm board: two MIPI-CSI camera connectors, a MIPI-DSI display output multiplexed with one of them, HDMI via a 4-lane LVDS converter, two CAN headers, an I2C and ADC header, two 20-pin GPIO headers, and three fan headers. Power comes over USB Type-C PD, though it wants a 20V/5A supply — worth planning for before the board arrives.
What Does the FRDM-IMX95-PRO Cost?
NXP lists it at $300, currently marked pending stock. Software is the i.MX Linux BSP with Yocto build support, plus NXP's GoPoint GUI for running edge AI demos out of the box across robotics, industrial networking, and vision use cases.
At that price it sits well above hobbyist boards and well below an industrial PC, which is the right position for what it is: a development platform for products that will ship. If you are weighing form factors for a smaller build, our Raspberry Pi versus mini PC comparison covers the tradeoffs at the other end of the scale, and the RAK7392 WisGate Connect shows what a Compute Module-based edge router looks like by comparison. More boards like this in our mini computer coverage.
Sources: CNX Software — August 11, 2026; NXP Semiconductors — August 11, 2026.
More Mini Computers Stories

PewCB One Etches Circuit Boards With a Desktop Laser
PewCB One is a 20W laser PCB fab that ablates 0.1mm traces in under an hour with no chemicals, priced at $5,000 with production from September 1.

PiShot Makes a Pocket Camera From a Raspberry Pi Zero
PiShot is an open-source 3D-printed camera built on a Raspberry Pi Zero 2 W, Camera Module 3, an ST7789 rear display, and a single 18650 cell.

ESPHome Starter Kit Puts Four Sensors on One $40 Board
The official $40 ESPHome Starter Kit pairs an ESP32-C6 board with four snap-on FPC sensor modules and needs no YAML editing to get an automation running.
