Intel launches new visual processing unit Myriad X to enhance the performance of artificial intelligence products

Movidius, a chip maker under Intel, which recently launched the world's first standalone AI accelerator in USB format, the Movidius Neural Compute Stick, announced the launch of the new Myriad X visual processing unit (VPU) on the 29th. This is the world's first system-on-chip (SoC) equipped with a dedicated neural network computing engine, which can be used for deep learning reasoning on the acceleration side, such as drones, robots, smart lenses, virtual reality and other products. Let's follow the mobile phone portable editor to learn about the relevant content.

It is understood that Myriad X introduces a new architecture called Neural Compute Engine, which is a chip DNN computing accelerator. Movidius said that with the help of this accelerator, Myriad X can achieve a maximum output of more than one trillion operations per second (TOPS) based on the theoretical computing power of 4+ TOPS. This can provide more enhanced performance under the same power consumption conditions for deep learning applications such as drones, robots, smart cameras, virtual reality (VR) and other products.

In addition, Myriad X has 4 vector processors with C language programming and 128-bit very long instruction set (VLIW). With the same MIPI channel as the previous generation Myriad 2 processor and the expanded 2.5 MB memory, Myriad X can achieve more image and visual accelerator functions. It is worth noting that, like Myriad 2, Myriad X's vector processor is also a dedicated SHAVE processor optimized for visual workloads in computers. Myriad X also supports the latest LPDDR4 memory.

Another new feature of Myriad X is 4K hardware encoding, which supports 4K hardware encoding at 30 Hz (H.264/H.265) and 60 Hz (M/JPEG). In terms of interface, Myriad X supports two external connection ports: USB 3.1 and PCIe 3.0. Everything will be done within 2W power, just like Myriad 2, and there is a chance that it will be lowered to 1W power.

The new Myriad X offers better power performance than the previous generation Myriad 2, thanks to the adoption of TSMC’s 16nm process. By shrinking from 28nm to 16nm FinFET, Movidius was able to save power on the Myriad X and invest more in the SHAVE processor, accelerator, interface, and memory.

Although Intel has its own fab, Movidius said that Myriad X was already under development before Intel acquired it in 2016, and TSMC's 16nm FinFET process has been selected for production. Compared with the previous Myriad 2 produced with a 28nm process, it is a full generation higher.

Although the exact details have not yet been announced, the Myriad X software development kit (SDK) was released together with a neural network compiler and a specialized FLIC framework for developing applications in a plug-in manner. Therefore, Myriad X, like Myriad 2, can be programmed through the exclusive Myriad Development Kit (MDK).

It is understood that Myriad 2 will not be replaced by Myriad X at this stage. This is because the price of Myriad 2 has dropped to less than $10 per chip in January 2016. Myriad X uses a more expensive 16nm FinFET process and additional hardware structure. Given its higher performance and likely higher price, it may not be easy to "change dynasties" in a short period of time.

The above is an introduction to Intel's launch of the new visual processing unit Myriad X in portable mobile phones to enhance the performance of artificial intelligence products. If you want to know more related information, please pay more attention to eeworld. eeworld Electronic Engineering will provide you with more complete, detailed and updated information.