Specifications — Virtium VMX-004 AI Accelerator Module
Manufacturer
Virtium (SOM platform with Embedded Artists)
Model / MPN
VMX-004
AI Processors
4× MemryX MX3 inference processors
AI Compute
Up to 24 TOPS
Form Factor
M.2 2280
Host Interface
PCIe Gen3, 2×2 lanes
Typical Power
3 – 10 W
Memory Architecture
Models run fully from on-chip SRAM — no external DRAM
Operating Temperature
−40 °C to +85 °C industrial
Frameworks
ONNX, PyTorch, TensorFlow,
Keras, TensorFlow Lite
OS Support
Ubuntu, Windows, Android
Host Compatibility
ARM and x86 carrier boards / motherboards
Compliance
Trade Agreements Act (TAA) compliant
Origin
Designed and assembled in Southern California, USA
Specifications — Virtium / Embedded Artists RZ/G3E DX-M1 SOM
Application Processor
Renesas RZ/G3E — 4× Cortex-A55 @ 1.8 GHz + Cortex-M33 @ 200 MHz
CPU Throughput
21.6K DMIPS total
AI Booster
DEEPX DX-M1 — up to 25 TOPS @ up to 1 GHz
AI Efficiency
25 TOPS at 5 W
On-chip NPU
Arm Ethos-U55, 256 MAC / 512 GOPS (0.5 TOPS) — integrated option
Accelerator Memory
4 GB LPDDR5-5600, 64-bit / 4-channel
System Memory
4 GB LPDDR4X-3200, 32-bit / 2-channel
Storage
64 GB eMMC NAND (OS + bootloader)
Accelerator Link
PCIe Gen3, 2 data lanes (DX-M1 to RZ/G3E)
Form Factor
EACOM / MXM3, 314 positions, 82 × 50 × 5 mm
Graphics
Mali-G52 @ 630 MHz, 30 GFLOPS, Vulkan 1.2 / OpenCL 2.0
Video
H.264 1080p60 & H.265 4Kp30 encode/decode
Displays
MIPI-DSI 4-lane, LVDS single/dual link, parallel RGB — dual display
Camera In
1× MIPI-CSI2, 4 lanes, up to 2.1 Gbps per lane
Networking
2× Gigabit Ethernet (Microchip KSZ9131 PHY)
Wireless (opt.)
Murata 2FY Wi-Fi 6E 802.11ax SISO + BT 5.4 (Infineon CYW55513)
USB
USB 3.2 Gen2 host-only + 2× USB 2.0
Supply / Power
4–5 V input; 7–9 W peak
Temperature
Industrial −40 to +85 °C / commercial 0 to 70 °C
Scalable Option
Available without DX-M1 mounted for Edge Compute only
Overview
Virtium is a 25-year industrial modular hardware manufacturer whose Edge AI line pairs its own manufacturing, qualification and lifecycle-management infrastructure with accelerator silicon from MemryX and DEEPX. The portfolio covers two complementary form factors: an M.2 accelerator card that upgrades an existing host, and a populated System-on-Module for new designs.
The Virtium VMX-004 is an Edge AI industrial module carrying four MemryX MX3 inference processors for up to 24 TOPS at 3–10 W typical. Its defining architectural choice is that models run entirely from on-chip SRAM, so the module needs no external DRAM — dropping latency, power, heat and board complexity at once. Floating-point activations preserve vision-model accuracy, and models compile with a one-click toolchain requiring no retraining or code changes; software access is available without an NDA. The card follows PCIe M.2 2280 and drops into most ARM and x86 carrier boards and motherboards.
The VMX-004 is a rare combination of edge AI acceleration with TAA compliance and US assembly — designed and assembled in Southern California — which makes it relevant to aerospace and defence, government and other procurement programmes that require domestic sourcing. Industrial −40 °C to +85 °C operation and long-term product availability discipline suit deployment outside climate-controlled racks.
The RZ/G3E DX-M1 SOM takes the opposite approach, integrating compute and acceleration on one module. A Renesas RZ/G3E application processor (4× Cortex-A55 at 1.8 GHz plus a Cortex-M33 real-time core, 21.6K DMIPS) is paired over PCIe Gen3 with a DEEPX DX-M1 AI Booster delivering up to 25 TOPS at 5 W, with its own 4 GB of LPDDR5-5600. Buyers who need only edge compute can order the same module without the DX-M1 mounted and scale into AI later without redesigning the carrier. Dual display, H.264 1080p60 / H.265 4Kp30 codecs, dual Gigabit Ethernet, optional Wi-Fi 6E and a MIPI-CSI2 camera input cover HMI, vision and multimedia roles, in the 82 × 50 mm MXM3 form factor.
Applications
Vision AI & Inspection
Smart manufacturing, machine vision, video analytics on streams
Aerospace & Defence
TAA-compliant, US-assembled sourcing for government programmes
Surveillance & Security
Multi-camera inference without a discrete GPU
Unmanned Systems
Drones and LEO satellite payloads where power and heat are constrained
Transportation & Agriculture
Industrial-temperature operation in vehicles and field equipment
Medical & HMI
Dual-display SOM for imaging panels and operator interfaces
Request a Quote — Virtium VMX-004 / RZ/G3E DX-M1
QS Compute — global B2B supply of AI computing hardware. Price on request, stock availability confirmed on inquiry.
Get Your Quote →