As Single Board Computers (SBC) are becoming increasingly popular to run AI and Deep learning some have even been specially designed just to run AI and Deep learning! Today, we are going to put 3 SBC up to a fight and compare their deep learning capabilities.
Let us introduce our 3 fighters for today!
- The Rock Pi N10. It is a new member of the Rock pi family that is born for AI and deep learning processing.
- It carries a powerful SoC(system on chip) which is RK3399Pro which features a CPU, GPU, and NPU.
- RK3399Pro’s CPU is a six-core CPU which includes Dual Cortex-A72(frequency 1.8GHz) and quad Cortex-A53(frequency 1.4GHz).
- The GPU of RK3399Pro is Mali T860MP4 which has the ability to support OpenGL ES 1.1 /2.0 /3.0 /3.1 /3.2, Vulkan 1.0, Open CL 1.1 1.2, DX1.
- As for NPU, the NPU can support 8/16 bit computing and up to 3.0 TOPS computing power.
- Rock Pi N10 also has plenty of resources for storage. 64 bits dual-channel 4GB LPDDR and 16GB eMMC 5.1 is embedded on the mainboard for providing enough storage for processing and saving data. Besides, the board also contains a μSD card slot for booting and even an M.2 SSD connector which supports up to 2T SSD for extending storage.
- The Rock Pi N10 is totally an interface monster. Like Raspberry 4B, Rock Pi N10 has rich interfaces for Audio, camera, display, Ethernet, USB and I/O pins. The Ethernet interface can support PoE function and has a PoE hat near the Ethernet interface. The SBC can not support wi-fi for now, but there will be an optional wi-fi module to be embedded on the board later soon
- The software system of this Rock Pi N10 board is Debian and Android 8.1. For the NPU, there is an upgraded firmware and booting procedure.
- Other boards that feature the RK3399Pro Rockchip like the Toybrick RK3399Pro AI Developer Kit costs above $200, but with the Rock Pi N10, it is more affordable starting at only $99.
- The Raspberry Pi 4 Model B is the latest product in the popular Raspberry Pi range of computers.
- It offers ground-breaking increases in processor speed, multimedia performance, memory, and connectivity while retaining backward compatibility and similar power consumption as the prior generation Raspberry Pi 3 Model B+.
- This product’s key features include a high-performance 64-bit quad-core processor, dual-display support at resolutions up to 4K via a pair of micro-HDMI ports, hardware video decode at up to 4Kp60, up to 4GB of RAM, dual-band 2.4/5.0 GHz wireless LAN, Bluetooth 5.0, Gigabit Ethernet, USB 3.0, and PoE capability (via a separate PoE HAT add-on).
- The dual-band wireless LAN and Bluetooth have modular compliance certification, allowing the board to be designed into end products with significantly reduced compliance testing, improving both cost and time to market.
- Due to the performance uplift of the Pi 4 compared to the previous model, the Raspberry Pi 4 is now a strong platform for on-device inference. In addition, with a USB 3.0 implemented, it makes the Pi 4 a strong host for AI and deep learning accelerator hardware
- The Raspberry Pi 4 is cost-effective as well at only $35 for 1GB RAM option up to $55 for 4GB of RAM.
- The NVIDIA® Jetson Nano™ Developer Kit delivers the compute performance to run modern AI workloads at unprecedented size, power, and cost. Developers, learners, and makers can now run AI frameworks and models for applications like image classification, object detection, segmentation, and speech processing.
- The developer kit can be powered by micro-USB and comes with extensive I/Os, ranging from GPIO to CSI. This makes it simple for developers to connect a diverse set of new sensors to enable a variety of AI applications. The Jetson Nano is also incredibly power-efficient, consuming as little as 5 watts.
- Jetson Nano features a Quad-core ARM® Cortex®-A57 MPCore processor, NVIDIA Maxwell™ architecture GPU with 128 NVIDIA CUDA® cores and 4 GB 64-bit LPDDR4 1600MHz memory.
- With this, makers are able to create devices with the Jetson Nano that can handle multiple machine learning tasks such as computer vision and natural language processing, all on a single, compact computer.
- It can also run multiple neural networks on each sensor streams and supports most of the most popular AI frameworks available today including TensorFlow, PyTorch, Caffe, and MXNet.
Rock Pi N10 RK3399Pro vs Raspberry Pi 4 vs Jetson Nano
When it comes to AI and machine learning, the CPU, GPU, and NPU play a very important part. For those who do not know what is a CPU, GPU or NPU:
- CPU: Known as Central Processing Unit, it is the core component and brains of the operation of the SBC. It takes instructions from a program or application and performs a calculation.
- GPU: Known as Graphics Processing Unit, is designed to rapidly manipulate and alter memory to accelerate the creation of images in a frame buffer intended for output to a display device. They are also used to accelerate other geometric calculations which are using in machine learning.
- NPU: Known as Neural Network Processing Unit is a specialized circuit that implements all necessary control and arithmetic logic necessary to execute machine learning algorithms.
Thus, we will compare their Specifications and look at their CPU, GPU, and NPU in particular:
|SBC||Rock Pi N10(Model A/B/C)||Raspberry Pi 4B||Jetson Nano|
|CPU||Dual Cortex-A72@ 1.8GHz and quad Cortex-A53 1.4GHz||Quad-core ARM Cortex-A72 64-bit @ 1.5 GHz||Quad-Core ARM Cortex-A57 64-bit @ 1.42 GHz|
|GPU||Mali T860MP4||Broadcom VideoCore VI (32-bit)||NVIDIA Maxwell w/ 128 CUDA cores @ 921 Mhz|
|NPU||3.0 TOPS computing power||–||–|
|LPDDR||4/6/8GB LPDDR3||4GB LPDDR4||4GB LPDDR4|
|Networking||Gigabit Ethernet only||Gigabit Ethernet / Wifi 802.11ac||Gigabit Ethernet / M.2 Key E (for Wifi support)|
|Display||HDMI 2.0||2x micro-HDMI (up to 4Kp60)||HDMI 2.0 and eDP 1.4|
|USB||1x USB 3.0, 2x USB 2.0||2x USB 3.0, 2x USB 2.0||4x USB 3.0, USB 2.0 Micro-B|
|Video Encoder||H264(1080p30) and VP8||H264(1080p30)||H.264/H.265 (4Kp30)|
VC-1, MPEG-1/2/4, VP8
|H.264/H.265 (4Kp60, 2x 4Kp30)|
|GPIO||40-pin GPIO||40-pin GPIO||40-pin GPIO|
The first thing you may realise is that Rock Pi N10 RK3399Pro is the only one with an NPU. With this NPU, the computing power of Rock Pi N10 is 3.0 TOPs, which means 3.0 x 10 ^ 12 times of data is processed per second!
As the NPU is not present on the Raspberry Pi 4 and Jetson Nano, they can only use their CPU to handle complex calculations for deep learning. This sole feature makes the Rock Pi N10 RK3399Pro superior compared to them in running AI and deep learning performance.
In terms of GPU, the Jetson Nano stands out in terms of their NVIDIA Maxwell w/ 128 CUDA cores @ 921 Mhz. As the RK3399Pro uses an integrated GPU together with its CPU and NPU, it loses out to the Jetson Nanon but it is just in terms of its AI image processing capabilities. The Raspberry Pi 4 GPU is also weaker when compared to the Jetson Nano.
As for CPU, the Raspberry Pi 4, has the newest and best CPU with its Quad-core ARM Cortex-A72 64-bit @ 1.5 GHz which provides a faster clock speed and performance. However, for deep learning and AI, it may not provide much performance benefits but with a faster CPU compared to the rest, the Raspberry Pi 4 is better suited to use as a general-purpose computer.
For the price component, as you can see, the Raspberry Pi 4 is the cheapest out of all of the SBCs. The Jetson Nano and the starting price of the Rock Pi N10 RK3399Pro are the same at $99.
However, do note that other boards that feature the RK3399Pro Rockchip like the Toybrick RK3399Pro AI Developer Kit costs above $200, but with the Rock Pi N10 RK3399Pro, it is much more affordable.
In summary, as you can see from the comparison, each SBC has its own advantages and disadvantages. However, for deep learning and AI purposes, the Rock Pi N10 RK3399Pro is still a beast when it comes to deep learning capabilities offering a powerful performance of up to 3.0 TOPS computing power due to its NPU. In addition, at its price point, you will definitely get your money worth with this SBC.
If your deep learning project involves heavy image and graphics processing, the Jetson Nano will be suitable with its NVIDIA Maxwell w/ 128 CUDA cores @ 921 Mhz.
If you are just looking to run basic deep learning and AI tasks like seeing movement, recognizing objects and basic inference tasks at a low FPS rate, the Raspberry Pi 4 would be suitable. If not, the Raspberry Pi 4 would still be better suited as a general-purpose computer.
What do you think? Which SBC won this fight? Do let us know in the comments section down below!
If you are interested and want to learn more about microcontrollers for AI and deep learning, you can check out our other blog here!