Atlas 300V Pro video parsing card

The Atlas 300V Pro video resolution card integrates "general-purpose processor, AI Core, and codec" into one, offering powerful AI inference and video image codec functions. It features a large number of video parsing channels, high-performance feature retrieval, and secure startup. It supports real-time analysis of 128 high-definition video channels, making it widely applicable in AI industry scenarios such as smart cities, smart transportation, smart parks, and smart finance.

1. Key Features & Specifications

• Core & Processing: Integrates a general-purpose processor, AI Core, and hardware codec, delivering 140 TOPS (INT8) and 70 TFLOPS (FP16) of AI compute.
• Memory: Equipped with 48 GB LPDDR4X memory (384-bit width, ECC support, and 204.8 GB/s bandwidth), doubling the memory capacity of the Atlas 300I Pro.
• Video Decoding & Encoding:
  • Decoding: Supports up to 128 channels of 1080p30 (H.264/H.265) or 16 channels of 4K60 video.
  • Encoding: Supports up to 24 channels of 1080p30 or 3 channels of 4K60 video.
  • JPEG: Supports 4K JPEG decoding at 512 FPS and encoding at 256 FPS (up to 8192x8192).
• Form Factor & Power: Standard HHHL (Half-Height Half-Length) PCIe Gen 4.0 x16 card with a 72W thermal design power (TDP).
• Virtualization: Supports up to 7 virtual NPUs.

2. Use Cases:

Strictly designed for AI inference jobs (does not support model training) in scenarios like smart cities, smart transportation, smart campuses, and finance. 

3. Technical specifications