From entry-level to high performance, how to select your AI devices?
Today the applications of artificial intelligence (AI) at the edge are expansive. It can be an AI traffic monitoring system for predicting traffic, preventing congestion and potential accidents, or an optical semiconductor wafer inspection system that incorporates AI to increase yield and quality. According to Tractica, AI edge device shipments will increase from 161.4 million units in 2018 to 2.6 billion units worldwide annually by 2025. Research and Markets also states that the global edge AI market is expected to grow at a compound annual growth rate of 19.27% over the forecast period to reach a market size of USD1,954.244 million in 2026 from USD 569.194 million in 2019.
In this article, we will introduce our edge AI systems from the light edge AI to the high-performance.
Light edge AI: With the accelerator module, AI can assist with work and life
In light edge AI applications, configuring with an edge system and an accelerator module is the most cost-effective way to deploy the applications, while maintaining low power consumption and maintaining good computing power. The accelerator usually comes as an M.2, mini PCIe, or MXM module. Combined with the use of cameras, it guards the workers with a warning when they don’t put on the personal protection equipment properly or enter the danger zone. Over the production line, it can prevent workers’ hands from reaching unsafe areas. In these kinds of applications, Axiomtek recommends its signature and hot selling eBOX series of fanless embedded systems.
With the support of the Hailo-8™ AI-accelerator, the eBOX series products now fully strengthen its competition from all scopes, maximizing the power of edge AI and helping customers speed up time to market.
eBOX630-528-FL
- 8th gen Intel® Core™ i7/i5/i3 or Celeron® ULT processor (Whiskey Lake-U)
- Dual-channel DDR4 2400 SO-DIMM up to 64GB of system memory
- Two PCIe Mini Card slots for Hailo-8™ AI mini PCIe acceleration modules or wireless modules
- Two swappable 2.5” HDD drive bays and one mSATA slot
- PoE function to support IP cameras for visual inspection (optional)
- A wide operating temperature range of -40°C to +70°C
eBOX626-311-FL
- Intel® Atom® x5-E3940 l DDR3L-1866 SO-DIMM for up to 8GB of system memory
- Two PCIe Mini Card slots for Hailo-8™ AI mini PCIe acceleration modules or wireless modules
- One 2.5” SATA HDD drive bay and one mSATA slot
- Four USB 3.0 and two USB2.0 to support cameras for visual inspection
- A wide operating temperature range of -40°C to +60°C
Advanced edge AI: With discrete graphics card or FPGA, the AI computing is brought to a more versatile level
In more advanced edge AI applications, a scalable edge system with a discrete graphics card should be able to handle more complicated situations, such as damage detection on the road, crop growth analysis or even autonomous mobile robot (AMR). If the budget is abundant and there is higher demand for acceleration, it can even add additional FPGA or VPU into the system. To faster image capture and processing right at the edge, it will be better if the system has higher scalability for a frame grabber. Axiomtek’s IPC series are scalable and high-performance systems that are reasonable choices for advanced edge AI. While the IPC960 series supports mini (half-size) GPU, the IPC970 series can even support full-size GPU.
Recently, a customer integrated the IPC970 into a vehicle for the AMR application. With the RTX 3090 GPU and the AX92321 multiple-port USB 3.0 frame grabber for the cameras, the IPC970 has been put into a self-driving baggage tractor at the airport. The automated tractor can go by predefined routes, management system orders or on-demand via the controller. It can detect and avoid unexpected obstacles thanks to a full set of sensors, dual Lidars, radars, cameras, a receiver, and measurement units. The IPC970 and RTX3090 process the information effortlessly from all the devices, ensuring the baggage and cargo go to the right airplane safely. As it operates day by day, the data collected in the process can re-train the operation gradually.
IPC970
- Intel® Xeon® or 10th Gen Intel® Core™ i7/i5/i3 processor
- Four DDR4-2933 ECC/non-ECC U-DIMM, up to 128GB of system memory
- Six USB 3.2 Gen.2, two USB 3.2 Gen.1 and one USB 2.0
- Two PCIe x8, one PCIe x16 for GPU
- Supports NVIDIA® GeForce RTX 3090 graphics card
- Two external SSD/HDD, one internal SSD/HDD drive bay
- One M.2 Key M 2280 slot for NVMe SSD
- One M.2 Key B 3042/3050 slot for 5G wireless connections
- One PCIe Mini Card slot and one M.2 Key E 2230 socket for Wi-Fi/LTE modules
- A wide operating temperature range of -10°C to +60°C
IPC962-525
- 9th/8th gen Intel® Core™ processor
- Two DDR4-2666/2400 un-buffered SO-DIMM, up to 64GB of system memory
- One PCIe x16, one PCIe x4 for GPU
- Supports NVIDIA® GPU card and NVIDIA® Tesla T4 GPU card
- Two 2.5" swappable HDD/SSD with the support of Intel® RAID 0,1
- One M.2 Key M 2280 slot for NVMe SSD
- One M.2 Key B 3050 slot for 5G wireless connections
- A wide operating temperature range of -10°C to +60°C
High-Performance AI Computing: With multiple GPU or FPGA, deep learning is within the reach
When it comes to sophisticated AI applications like deep learning and video analysis, it will be much more demanding for the GPU workstation. In addition to a high-end processor, the advanced GPU workstation should be equipped with multiple PCIe slots and more multiple GPU and FPGA to push the performance to a higher level.
The system can process an unimaginably heavy load of data; it plays the role of a GPU workstation; while it receives data from the edge systems, it can retrain on data and keep learning fast and persistently then renew models and send them back to the edge to improve the processing and accuracy. Having a more sophisticated system means better capability; it is used in hospitals for AI disease detection and even more complicated wafer inspection in the semiconductor industry. In both cases, the AI model must be updated in real-time for better inference and more precise results.
Axiomtek’s GPU workstation, iHPC300, has been used in wafer inspection in the semiconductor industry. High-value wafers are not only fragile; most difficult of all, the defects are hair-like which is likely to lead to over kill or under kill. Configured with inspection microscopes, the model training server has read over 500,000 images before completing building a deep learning model. The AI wafer inspection solution resulted in automation and better quality.
iHPC300
- LGA4189 socket Intel® Xeon® Scalable processor, up to 270W, 40 cores
- Six DDR4-3200 R-DIMM un-buffered memory, up to 384GB of system memory
- Three PCIe x16 and three PCIe x8 slots, support up to six accelerator cards
- Six SATA3 with RAID 0/1/5/10, exposed three 5.25" HDD and one 3.5” HDD drive bays
- Three internal 3.5" HDD drive bays
- One M.2 Key M2280 for NVMe SSD
- Supports 2000W/1200W power supply
