Intel unveils tiny Myriad X AI chip to run deep neural networks at high speed and low power
Silicon Valley has been going guns-a-blazing on artificial intelligence for some years now, but it’s still very much the buzz word in the tech industry. Intel seems to agree. In a bid to bolster its position in the AI landscape, the chip giant has just unveiled the Myriad X, a more powerful version of the Myriad 2 computer vision-minded chip built for powerful imaging applications.
Built by Movidius, a specialist IoT firm that was acquired by Intel almost a year ago, the Myriad X is apparently the world’s first system-on-chip to feature a dedicated Neural Compute Engine. That is, an on-chip hardware block specifically designed to run deep neural networks at high speed and low power while retaining good accuracy, enabling devices to see, understand and respond to their environments in real time.
Intel swallowed California-based chip outfit Movidius in September to improve robots’ awareness of the real world. By buying the firm, which specialises in low-power-chip design for machine intelligence algorithms, Intel hoped to enhance its RealSense human-computer interaction camera technology, which provides face, gesture and speech recognition, as well as augmented-reality capabilities.
“With the introduction of the Neural Compute Engine, the Myriad X architecture is capable of 1 TOPS of compute performance on deep neural network inferences,” Intel said in its announcement. And if you’re wondering what TOPS is, it refers to a trillion operations per second (TOPS).
“We’re on the cusp of computer vision and deep learning becoming standard requirements for the billions of devices surrounding us every day,” explained Movidius’ vice president and general manager, Remi El-Ouazzane.
“Enabling devices with humanlike visual intelligence represents the next leap forward in computing. With Myriad X, we are redefining what a VPU means when it comes to delivering as much AI and vision compute power possible, all within the unique energy and thermal constraints of modern untethered devices.”
El-Ouazzane added that the Myriad X’s tiny form factor and on-board processing makes it ideal for autonomous device solutions, such as driverless cars, and in addition to its Neural Compute Engine, Myriad X is able to combine imaging, visual processing and deep learning inference in real time thanks to four main features.
The first of which is a “programmable 128-bit VLIW vector processor”, which Intel says is capable of running multiple imaging and vision application pipelines simultaneously, as well as having the flexibility of 16 vector processors built for computer vision workloads.
Then there’s “increased configurable Mobile Industry Processor Interface (MIPI) Lanes”, which means the chip can connect up to 8 HD resolution RGB cameras directly to Myriad X. This, alongside its 16 MIPI lanes, means it can support up to 700 million pixels per second of image signal processing.
Another new feature in the Myriad X is “enhanced vision accelerators”, which are able to make use of over 20 hardware accelerator, performing tasks such as optical flow and stereo depth without introducing additional compute overhead.
The final major feature Intel is touting on its new Myriad X chip is its 2.5 MB of homogenous on-chip memory. This centralised memory architecture allows for up to 450 GB per second of internal bandwidth, which Intel claims reduces latency and power consumption by minimising off-chip data transfer.
Intel hasn’t released details on pricing for the Myriad X, but with all these new features, it’s most likely to be a pricier option over the Myriad 2, which Intel said it will not replace.