AI Accelerator Card - Cambricon Technologies

Cambricon Technologies

The newly upgraded Cambricon basic software platform has added the reasoning acceleration engine MagicMind, achieving integrated training and promotion. This significantly enhances the efficiency of development and deployment, and reduces users' learning,

Book a Demo

AI Accelerator Card

With fast deployment and centralized management, achieve simplicity at speed and scale with RangeFar hyperconverged infrastructure. Save over 60% compared to VMware and lower your TCO.

Siyuan 370 chip

Based on the 7nm manufacturing process, the chiplet 370 is Cambricon's first AI chip to adopt the Chiplet technology, integrating 39 billion transistors and having a maximum computing power of up to 256TOPS(INT8), which is twice the computing power of Cambricon's second-generation product, the Chiplet 270.

Siyuan 270 Series

The Siyuan 270 adopts the Cambricon MLUv02 architecture and can support a variety of artificial intelligence applications such as vision, speech, natural language processing, and traditional machine learning. It also integrates abundant video and image codec hardware units for vision applications.

Feature To connect the world with high-value, globally proven Chinese technology — bridging innovation with real-world application through trusted, scalable solutions.

MagicMind Inference Acceleration Engine The industry's first inference engine that has achieved commercial deployment capabilities based on MLIR graph compilation technology. Users only need to invest a very small amount of development costs to deploy inference services to the entire Cambricon product line and obtain highly competitive performance.

An integrated training and promotion software development platform Cambricon's basic software platform integrates the entire underlying software stack for training and inference, including underlying drivers, runtime libraries, operator libraries, and toolchains, etc. It deeply fuses MagicMind with the artificial intelligence frameworks Tensorflow and Pytorch to achieve an integrated training and inference system.

Low-power high-bandwidth LPDDR5 memory The Siyuan 370 chip is the first in the industry to support LPDDR5 memory, featuring high bandwidth and low power consumption. Its memory bandwidth is three times that of the previous generation, and its memory access efficiency is 1.5 times that of GDDR6. It can allocate more energy to AI chips within the limited power consumption range of the board card and output more powerful computing power.

H3C

Inspur

Cambricon Technologies

Paraview

Sangfor Technologies

ArcherOS

ShareCRM

Cambricon Technologies