Flex Logix InferX X1M M.2 AI accelerator module is now in production

When we first wrote about Flex Logix InferX X1 AI inference accelerator in 2020, the InfexX X1M M.2 AI accelerator module for edge servers, PCs, laptops, and embedded systems. was expected in Q2 2021.

It took longer than planned but the company has finally announced production availability of the InferX X1M board in M.2 2280 form factor, or about the size of a stick of gum, to fit into power-constrained applications such as robotic vision, industrial, security, and retail analytics.

InferX X1M

InferX X1M module specifications:

  • InferX X1 Edge Inference Accelerator with 4K MACs with 8 MB of distributed SRAM, and an eFPGA with 4MB DDR3 RAM
  • System Memory – LPDDR4X (capacity undisclosed)
  • Interface – M.2 B+M Key edge connector with x4 PCIe GEN3/4
  • Dimensions – 22mm x 80mm
  • Power Budget – 6 to 8.2W TDP

Flex Logix does not provide a lot of information on the product page, and instead, it’s better to check out a 2021 presentation to get more details, notably about performance.

InferX X1M benchmarks yolov3

The cards can achieve 30 inferences per second (IPS) at 6W on YOLOv3 with 416×416 image and a 6W TPD, and the number goes up to a little over 50 IPS at 8.2W TDP. Back in 2020, the company expected to run the X1 with a power budget of 19W on the M.2 card, but this would clearly not be possible due to the thermal limitations of the M.2 form factor that can handle up to around 8W.

The company further says the InferX X1M board can handle Yolov5 and is optimized for large models and megapixel images at batch=1 for object detection and other high-resolution image processing capabilities needed for edge servers and industrial vision systems. Flex Logix also provides software tools that include a utility to port trained ONNX models to run on the X1M, a simple runtime framework to support inference processing within both Linux and Windows, and a driver with external APIs to configure & deploy models.

X1 software tools

As far as I can tell, none of those tools are available publicly so you’d have to purchase the module and ask the company to provide them. Flex Logix InferX X1M M.2 AI accelerator board is supposed to sell for $399 per unit for 1K orders.

Thanks to TLS for the tip.

Share this:

Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress

ROCK Pi 4C Plus
Notify of
The comment form collects your name, email and content to allow us keep track of the comments placed on the website. Please read and accept our website Terms and Privacy Policy to post a comment.
Khadas VIM4 SBC