256TOPS
Up to 256TOPS physical computing power @INT8
35W
35W Typical Dissipation
Supportexternal expansion ofMemory
with a bandwidth of128GB/s
SupportPCle 4.0
SupportEP mode
256TOPS
Up to 256TOPS physical computing power @INT8
35W
35W Typical Dissipation
Supportexternal expansion ofMemory
with a bandwidth of128GB/s
SupportPCle 4.0
SupportEP mode
-
Superior AI Computing power
The physical computing power of chips reaches 256TOPS, which significantly enhances the interaction efficiency between data and computing units, and the computing utilization rate is extremely high
-
High energy efficiency ratio
The core AI processor has an energy efficiency ratio of up to 20TOPS/Watt, far ahead of traditional architectures
-
Low latency
The latency of AI computing data processing is reduced by more than 2 times compared with traditional AI chips, being suitable for various latency-sensitive applications
-
User agile development
We provide a powerful and easy-to-use tool chain to support TensorFlow, PyTorch, ONNX and third-party custom frameworks, and significantly enhance development efficiency
