Movidius has been making some waves in this field recently. In January, the startup inked a deal with Google to assist with its efforts to improve image recognition on mobile devices, and its impressive ‘visual processing unit’ intended for use in VR headsets garnered some buzz just a few weeks earlier.
Now, the Fathom looks set to facilitate all kinds of neural networking projects. Plug the stick in via a USB 3.0 port, and you can expect 20 to 30 times better performance in neural computing tasks — perfect for the prototyping process, before more powerful hardware is used to run the finished product.
The Fathom is built around a Myriad 2 MA2450 VPR and 512MB of LPDDR3 RAM, according to a report from Engadget. The all-important VPU is specifically designed for applications relating to neural networks; its architecture is far removed from standard CPUs and GPUs, allowing it to process up to 150 billion floating operations per second on no more than 1.2 watts of power.
The Fathom offers significant utility to researchers prototyping neural networks, and there’s plenty of scope for the device to be adopted by tinkerers and hobbyists.
“A prime example might be a company that makes relatively low volume drones for agriculture, heavy industry, or GIS,” Marketing Communications Director Jack Dashwood told Digital Trends. “The Fathom Neural Compute Stick could be plugged into these machines without extensive hardware replumbing and run neural network based applications such as object classification, scene labelling, etc.”
“Another example would be connecting this into a Raspberry Pi based robot, or home automation system, allowing some interesting automation functions like triggering door locks if it detects nobody home.”
Movidius is shipping Fathom Neural Compute Sticks to certain organizations right now, but the general public will have to wait a little longer. The good news is, the device won’t break the bank — when it launches in the winter, it’s expected to cost less than $100.
Updated 4-28-2016 by Brad Jones: Updated to remove an erroneous statement and to add comment from the company. Movidius is not backed by Alphabet.
Editors' Recommendations
- Deep learning A.I. can imitate the distortion effects of iconic guitar gods
- Japanese researchers use deep learning A.I. to get driftwood robots moving
