HUAWEI Ascend Atlas 300I Duo Inference Card

HUAWEI Ascend Atlas 300I Duo inference card

The Atlas 300I Duo inference card provides high- performance AI inference and video analysis thanks to its superior computing power and ultra-high energy efficiency. By integrating the general-purpose processor, AI Core, and codec, the 300I Duo offers premium speech and video analysis, retrieval clustering, content moderation, and OCR, for Internet, smart city, and smart transportation scenarios.

Specifications of the Atlas 300I Duo Inference Card

Feature	Specifications
Form factor	FHFL single-slot (10.5 inches)
AI processor	2 x Ascend 310P AI Processors, including: ● 16 Da Vinci AI Cores ● 16 Huawei-developed CPU cores
Memory	● LPDDR4X ● Capacity: 48 GB ● Total bandwidth: 408 GB/s ● Error checking and correcting (ECC)
AI computing powera	● Half-precision (FP16): 140 TFLOPS ● Integer precision (INT8): 280 TOPS
Encoding/Decoding capability	● H.264/H.265 decoder hardware decoding, 256-channel 1080p 30 FPS or 32-channel 4K 60 FPS ● H.264/H.265 encoder hardware encoding, 48-channel 1080p 30 FPS or 6-channel 4K 60 FPS ● JPEG decoding capability of 4K 1024 FPS and encoding capability of 4K 512 FPS, maximum resolution of 8192 x 8192
Virtual instance specifications	One Ascend AI Processor can be divided into several virtual NPUs in virtualization mode. Each virtual NPU supports 1, 2, or 4 AI Cores, and other hardware resources (such as memory and codec module) are divided proportionally.
PCIe port	● x16 lanes, compatible with x8/x4/x2 ● PCIe Gen4.0, compatible with 3.0/2.0/1.0
PCI IDs	Vendor ID: 0x19E5 Device ID: 0xD500 Subsystem vendor ID: 0x0200 Subsystem Device ID: 0x0110
Board Power	Maximum power consumption: 150 W
Dimensions (W x H x D)	266.7 mm × 18.46 mm × 111.15 mm (10.5 in. x 0.72 in. x 4.37 in.)
Weight (g)	910 g