2024 Ebook
Snapshots Into The Edge
Current Trends For Edge Computing
& How Premio Is Positioned
13th Gen Intel® Core™ Processors
Industrial x86 Rugged Computers
13th Gen Intel® Core™ Processor Industrial Computers
Edge AI Acceleration
with Hailo-8™ M.2 Modules
Premio is part of Hailo’s hardware partner ecosystem that enables technology leaders to bring advanced deep learning capabilities to a variety of edge computers with domain-specific performance acceleration. Premio has tested, validated, and benchmarked Hailo-8™ M.2 Modules in our rugged edge computers. Read Press Release ...
M.2 Form Factor
EDGE AI Performance
Power Efficent
Scalability
Real-time Performance
Modular EDGEBoost I/Os: Scalable AI Modules
Our EDGEBoost I/O Modules are purpose-built to support Hailo-8 acceleration modules for maximum Edge AI performance. Premio’s modular and rugged design enables seamless compatibility with our rugged edge computers for ultimate scalability.
- Seamless integration, scalability, and compatibility with EDGEboost I/O Modules
- Ruggedized and Passive Cooling design for harsh environments
- Scalable up to 4x Hailo AI Accelerators - 104 TOPS
- Industry-leading performance and number of Hailo-8 modules
- High-Speed PCIe for low-latency and fast performance.
RCO Series: Compatible with Hailo-8™ M.2 AI Accelerators
Premio’s flagship line of industrial computers seamlessly integrates Hailo-8 AI Accelerators to enable real-time AI inferencing at the rugged edge. This line of industrial computers, the RCO series, is purpose-built with EDGEBoost technologies to deliver tailored edge computing solutions for high specification Industry 4.0 applications.
RCO-1000-EHL Ultra-Compact Mini Computer
Palm-sized industrial computer for extremely space-constrained deployments.
• Supports 1x Hailo-8™ M.2 AI Accelerator (26 TOPS)
• On-board (1x M.2 B-Key, PCIe x1)
• Product Performance Benchmarks
| ResNet-50 v1 | 1332 | 1332 |
| MobileNet_v2_1.0 | 2444 | 2444 |
| EfficientNet_M | 889 | 889 |
| SSD_MobileNet_v1 | 1055 | 1055 |
| YOLOv5m | 218 | 218 |
| Stdc1 | 54 | 22 |
| YOLOv3 | 69 | 37 |
|
Hailo Intel® Core™ i5-9400 |
RCO-1000-EHL Intel® Celeron® J6413 |
RCO-3000-CML Small Form Factor Computer
Multi-core processing with minimal footprint for space-limited edge AI deployments.
• Scalable up to 3x Hailo-8™ M.2 AI Accelerators (78 TOPS)
• On-board (1x M.2 B-Key, PCIe x2)
• EDGEBoost I/O Module (2x M.2 B-Key, PCIe x2)
• Product Performance Benchmarks
| ResNet-50 v1 | 1332 | 3996 |
| MobileNet_v2_1.0 | 2444 | 7333 |
| EfficientNet_M | 889 | 2672 |
| SSD_MobileNet_v1 | 1055 | 3166 |
| YOLOv5m | 218 | 654 |
| Stdc1 | 54 | 95 |
| YOLOv3 | 69 | 185 |
|
Hailo Intel® Core™ i5-9400 |
RCO-3000-CML 10th Gen Intel® Core™ Processor |
RCO-6000-CML AI Edge Inference Computer
Utilizes modular EDGEBoost Nodes with GPU and NVMe acceleration for maximized edge performance.
• Scalable up to 4x Hailo-8™ M.2 AI Accelerators (104 TOPS)
• 2x EDGEBoost I/O Modules (2x M.2 B-Key, PCIe x2)
• Product Performance Benchmarks
| ResNet-50 v1 | 1332 | 5328 |
| MobileNet_v2_1.0 | 2444 | 9778 |
| EfficientNet_M | 889 | 3563 |
| SSD_MobileNet_v1 | 1055 | 4222 |
| YOLOv5m | 218 | 872 |
| Stdc1 | 54 | 95 |
| YOLOv3 | 69 | 221 |
|
Hailo Intel® Core™ i5-9400 |
RCO-6000-CML 10th Gen Intel® Core™ Processor |
RCO-6000-RPL AI Edge Inference Computer
Utilizes modular EDGEBoost Nodes with GPU and NVMe acceleration for maximized edge performance.
• Scalable up to 3x Hailo-8™ M.2 AI Accelerators (78 TOPS)
• 2x EDGEBoost I/O Modules (3x M.2 B-Key, PCIe x2)
• Product Performance Benchmarks
| ResNet-50 v1 | 1332 | 3996 |
| MobileNet_v2_1.0 | 2444 | 7333 |
| EfficientNet_M | 889 | 2672 |
| SSD_MobileNet_v1 | 1055 | 3166 |
| YOLOv5m | 218 | 654 |
| Stdc1 | 54 | 95 |
| YOLOv3 | 69 | 185 |
|
Hailo Intel® Core™ i5-9400 |
RCO-6000-RPL 13th Gen Intel® Core™ Processor |
| RCO-1000-EHL | RCO_3000-CML | RCO_6000-CML | RCO_6000-RPL | ||||||||||||||
| ONBOARD - 1x M.2 (B Key, 2242/3042/3052, PCIex 1) | ON-BOARD 1x M.2 B Key, 2242/3042/3052 (PCIex2, Support AI Module/NVMe Storage) + EBIO-2M2BK EDGEBoost I/O Module With 2x M.2 B-Key | EBIO-2M2BK EDGEBoost I/O Module With 2x M.2 B-Key | EBIO-2M2BK EDGEBoost I/O Module With 2x M.2 B-Key | ||||||||||||||
| NN Models | Resolution | FPS | Power [W] | FPS/W | % | FPS | Power [W] | FPS/W | % | FPS | Power [W] | FPS/W | % | FPS | Power [W] | FPS/W | % |
| Classification | |||||||||||||||||
| ResNet-50 v1 | 224x224 | 1,331.75 | 3.45 | 386 | 99.98% | 3,996.50 | 3.82 | 1,046 | 300.04% | 5,328.64 | 3.75 | 1,421 | 400.05% | 3996.22 | 3.69 | 361.45 | 300.02% |
| MobileNet_v2_1.0 | 224x224 | 2,444.47 | 2.15 | 1,136 | 100.02% | 7,333.66 | 2.24 | 3,274 | 300.07% | 9,778.16 | 2.20 | 4,445 | 400.09% | 7333.59 | 2.13 | 1149.31 | 300.07% |
| EfficientNet_M | 240x240 | 890.83 | 3.50 | 255 | 100.21% | 2,672.55 | 3.11 | 859 | 300.62% | 3,563.41 | 2.49 | 1,431 | 400.83% | 2672.57 | 3.78 | 235.70 | 300.63% |
| Object Detection | |||||||||||||||||
| SSD_MobileNet_v1 | 300x300 | 1,055.70 | 2.20 | 480 | 100.07% | 3,166.70 | 2.48 | 1,277 | 300.16% | 4,222.32 | 2.46 | 1,716 | 400.22% | 3167.51 | 2.43 | 434.65 | 300.24% |
| YOLOv5m | 640x640 | 218.10 | 4.50 | 48 | 100.05% | 654.16 | 5.10 | 128 | 300.07% | 872.24 | 5.01 | 174 | 400.11% | 654.19 | 5.19 | 41.98 | 300.09% |
| Segmentation | |||||||||||||||||
| stdc1 | 1024x1920 | 22.60 | 2.90 | 8 | 41.85% | 95.00 | 2.16 | 44 | 175.92% | 95.06 | 1.82 | 52 | 176.04% | 114.66 | 2.32 | 13.08 | 212.34% |
| Multi Stream Object Detection (8 Streams) |
|||||||||||||||||
| YOLOv3 | 608x608 | 37.86 | 4.90 | 8 | 54.87% | 185.57 | 4.91 | 37.79 | 268.94% | 221.00 | 4.42 | 50 | 320.29% | 198.46 | 4.28 | 46.36 | 287.62% |
| Average | 85.29% | 277.97% | 356.80% | 285.86% | |||||||||||||
Notes:
* RCO-1000-EHL CPU Host benchmark: Intel® Celeron® J6413
* RCO-3000-CML CPU Host benchmark: Intel® Core™ i7-10700TE
* RCO-6000-CML CPU Host benchmark: Intel® Core™ i7-10700TE
* RCO-6000-RPL CPU Host benchmark: Intel® Core™ i7-13700TE
* Power [W] listed is averages of single or multiple Hailo-8 module only, contact us for overall system power consumption
* FPS/W is based on total FPS and avg power on single or multiple Hailo-8 modules
Benefits of Hailo-8™ M.2 AI Acceleration Modules
Achieve accurate real-time edge AI operations without having size, power, or temperature constraints. The Hailo-8™ M.2 AI accelerators deliver enterprise-grade performance with 26 tera operations per second (TOPS) while operating at an efficient 2.5W. Each Hailo-8™ AI accelerator offers unmatched scalability with 100% utilization rates and linear performance stacking.
- Real-time AI Inferencing with 26 TOPS
- 2.5W Power Efficiency
- Fast time-to-market with standard M.2 form factor
- Scalable with linear performance stacking
- Support for standard AI frameworks
Hailo in Action: Webinar
Gain access to our exclusive webinar session featuring experts from Hailo and Premio to elaborate on the latest trends and challenges in security and surveillance.
Achieving Real-time Edge AI Inferencing
In this whitepaper, we introduce the benefits of domain-specific architectures and its capability to enable real-time AI processing with significantly reduced latency and bandwidth consumption, while maintaining low-power efficiency.
Edge AI Inferencing Deployment Applications
Industrial Automation
Increase production line efficiencies with AI-powered defect detection. Accurately identify and remove defected components, eliminating risk of human error. Minimize downtime with optimized maintenance strategies using predictive maintenance.
- Automated Quality Control: rapidly inspect and detect defective components
- Predictive Maintenance: monitor and ensure equipment operability with optimized maintenance planning
Security & Surveillance
24/7 monitoring and surveying critical sites to alert dispatch of early stages of potential hazards. Significantly amplifying effectiveness, productivity, and overall premise safety.
- Hazard Detection: identify initial stages of hazards and alert dispatch
- Safety Monitoring: ensuring safety standards and protocols are met
Transportation
Analyze and record traffic patterns for effective data-driven traffic control. Modernize tolling systems to allow for convenient non-stop highway tolling and web payment.
- Real-time Telematics: accurate vehicle positioning data for traffic control and analysis
- Automated Tolling: capture vehicle identification on highways
Certified by UL Solutions
For additional re-assurance in product quality and safety, this edge computer has earned the UL Certified Mark for ultimate reliability when operating in extreme industrial deployments.
• UL Listed (I.T.E E357184)
Why Premio?
Built Rugged. Built Ready.
- • 30+ years of extensive design expertise for industrial computing solutions in x86 compute power, storage, rich I/O, and high-speed connectivity
- • Global turnkey manufacturing and support infrastructure in the USA
- • Regulatory testing and compliance for rugged industrial computers
- • Long Product Life Cycles to ensure reliability
Related Articles
FAQ
M.2 AI Acceleration is a domain-specific architecture specifically to process AI workloads. M.2 is the form factor and how the processor is integrated into Premio’s rugged edge computers.
AI Accelerators efficiently processes data-intensive AI workloads with low-power consumption and within a compact form factor. This allows AI Accelerators to power AI algorithms, like machine/computer vision, in confined industrial spaces and provide seamless scalability for even greater workloads.
M.2 AI Acceleration Modules such as the Hailo-8 have a wide operating temperature range to be implemented into industrial computers and confidently process AI workloads in harsh industrial environments. Premio has developed EDGEBoost I/O Modules for seamless integration and dedicated heatsinks for the Hailo-8.
M.2 AI Accelerators are installed onto the on-board M.2 slot or on our EDGEBoost I/O Modules. Premio’s EDGEBoost I/O Modules feature an M.2 carrier board variation to house up to two M.2 AI accelerators on PCIe Gen 3.0 interface.
The RCO series is scalable to support up to 4x Hailo-8 AI Accelerators. The RCO-1000-EHL is only compatible with 1x Hailo, while the RCO-3000-CML can support up to 3x, and RCO-6000-CML up to 4x.