About VPU

顾名思义， Movidius 把一个面向机器视觉调优的深度学习神经网络框架缩小到了一个 U 盘大小，就变成了 Fathom。开发者可以直接把 Fathom 插到他们的硬件上，不再需要“借用”机载的处理器或者增加新的处理器。比如用在无人机上，无人机从传感器、摄像头获取的图像传输到 Fathom 计算棒里，而计算棒能以极低的功率和极高的效率（16-20 图像/秒/瓦）向无人机传回元数据（人、物体、障碍等等）

驱动 Fathom 的核心是 Movidius 前年推出的第二代 VPU：Myriad 2。

如果说摄像机、传感器是计算机视觉的“眼睛”的话，VPU 的定位就是的“大脑”。与适合多种用途的 CPU、GPU 不同，VPU 专门为计算机视觉进行优化，可以用于 3D 扫描建模、室内导航、360°全景视频等更前沿的计算机视觉用途。相较主流的移动处理芯片（集成 GPU 的 SoC），Myriad VPU 的身材更小，视觉处理运算的效能更高。

Myriad 2 采用 28 纳米制程，每秒可以进行 3 万亿次浮点运算，功耗不到 0.5 瓦，同时支持 6 个 60 帧 1080P 视频内容输入。此前接受采访时，该公司CEO Remi El-Ouazzane 对 PingWest品玩记者透露，Myriad 2 跟其他用于进行机器视觉处理的“非专业处理器”相比，能达到“1/10 的功耗、1/5 的尺寸和 1/5 的价格，10 倍的视觉处理性能。”

Fathom 适配性很广，可以支持 PC、无人机、机器人等硬件平台。它可以在不到 1 瓦的工作功率下运行一个经过完全训练的神经网络。它的学习能力可以用于物体识别、自然语言理解、自动驾驶等。

The VPU’s 10 SHAVE 128-bit SIMD vector units should be rather capable, easily beating Nvidia Tegra hardware when it comes to visual processing applications. Interestingly, the chip uses shared memory between the vector units and on board RISC CPUs. There’s also only 2MB of it, which is far less than many more contemporary computational systems employ.

The design principles for Myriad 2 followed from a number of improvements including an increase in the number of programmable vector-processors, and additional dedicated hardware accelerators. As a Vision Processing Unit (VPU) System-on-Chip (SoC), Myriad 2 has a software-controlled, multi-core, multi-ported memory subsystem and caches which can be configured to allow a large range of workloads. Myriad 2 provides exceptionally high sustainable on-chip data and instruction bandwidth to support the twelve processors, 2 RISC processors and high-performance video hardware accelerators.

In order to guarantee sustained high performance and minimize power, the Movidius proprietary processor called SHAVE (Streaming Hybrid Architecture Vector Engine) contains wide and deep register-files coupled with a Variable-Length Long Instruction-Word (VLLIW) controlling multiple functional units including extensive SIMD capability for high parallelism and throughput at both a functional unit and processor level. The SHAVE processor is a hybrid stream processor architecture combining the best features of GPUs, DSPs and RISC with both 8/16/32 bit integer and 16/32 bit floating point arithmetic as well as unique features such as hardware support for sparse data structures. The architecture is designed to maximize performance-per-watt while maintaining ease of programmability, especially in terms of support for design and porting of multicore software applications.