APEX-A100

APEX-A100 is a compact, high-performance edge AI computing platform built for demanding edge environments. With a fanless design, -40°C to 70°C operating range, and shock and vibration resistance, it ensures reliable 24/7 operation, while its 180 x 108.65 x 69 mm chassis enables easy integration across diverse edge AI applications and deployment scenarios.

Offering up to 100 TOPS (Dense) / 200 TOPS (Sparse) of AI computing, the product is tested and verified in real-world workloads, delivering excellent inference speed and performance across vision computing AI models and generative models.

Note 1:
 

1. Both solutions are configured with power consumption below the 30W standard
2. The other solution is set to Mode_30W
3. The Qualcomm solution is set to htp_performance_mode: 2
4. Input resolution: YOLO Series – 640×640; UNet-Seg – 640×1280; ResNet-50 – 224×224
    

Note 2:

• Dense computation: A reliable and straightforward metric representing the peak compute capability of a processing unit for neural network workloads.
   
• Sparse computation: Achieves higher effective compute performance (TOPS) and efficiency by skipping zeros or redundant weights. However, it requires careful model design to minimize potential accuracy loss.

Introducing IQ Studio — Innodisk‘s quick-start portal that empowers AI applications and accelerates edge deployment. IQ Studio serves as a one-stop platform providing BSPs, AI sample applications, benchmarking, and testing tools—all within an integrated development environment.

Streamline workflows, reduce setup time, and drive the rollout of edge AI solutions.

  Explore on GitHub > 

The Qualcomm Dragonwing™-powered COM-HPC Mini is where edge AI begins — and with Innodisk’s proven, fully compatible expansion options, we enable diverse applications and deliver a faster, simpler path to deployment.

Innodisk‘s camera modules bring AI vision to life, while high-performance SSDs and full-spectrum I/O expansion cards can be tailored to meet the unique needs of any AI application.
 

Qualcomm AI Hub offers over 100 validated AI models for Qualcomm edge devices, covering vision, speech, and generative AI applications such as YOLO, ResNet, and UNetSeg. Developers can easily find suitable AI models by selecting target CPU, test performance, and deploy them to edge devices within minutes.

  Explore on GitHub >     Explore Qualcomm AI Hub > 

The platform supports TFLite, QNN, and ONNX, fully leveraging Qualcomm’s heterogeneous computing architecture (NPU / CPU / GPU) for high performance and low power consumption at the edge.