Huawei 300i Pro Duo High-performance Al Inference Card

The Atlas 300I Pro inference card integrates "general-purpose processor, AI core, and codec" into one, offering powerful AI inference and target retrieval functions. It boasts advantages such as powerful computing power, ultra-high energy efficiency, high-performance feature retrieval, and secure startup, making it widely applicable in many AI scenarios such as OCR recognition, speech analysis, search recommendation, and content moderation.

1. Key Specifications

• AI Processor: 1 x Ascend 310P series processor (8 Da Vinci AI Cores, 8 Huawei-developed CPU Cores)
• AI Computing Power: 140 TOPS (INT8) / 70 TFLOPS (FP16)
• Memory: 24 GB LPDDR4X with ECC, 384-bit width, 4266 Mbit/s rate, and 204.8 GB/s total bandwidth
• Host CPU: 8 cores @ 1.9 GHz
• Video/Image Codec: Hardware decoding/encoding for H.264, H.265, and JPEG
• Interface: PCIe 4.0 x16 (backward compatible with Gen 3.0/2.0/1.0)
• Form Factor & Power: HHHL PCIe card, 169.5 mm x 68.9 mm, 72 W board power, 280 g weight
• Virtualization: Can be partitioned into up to 7 virtual NPUs

2. Common Use Cases

The Atlas 300I Pro is engineered primarily for high-density edge and data center video analysis (supporting up to 256 channels of real-time HD video), OCR recognition, speech analysis, search recommendations, and content moderation. It is distinct from the dual-chip Atlas 300I Duo model and relies on Huawei’s proprietary CANN software ecosystem rather than NVIDIA CUDA.

3. Technical specifications