Skip to main content

Huawei Ascend 510

Product Overview​

Huawei Ascend 510 is an AI chip in Huawei's Ascend series for edge computing and on-device inference, positioned as the edge-inference workhorse above the Ascend 310P. According to public selection data, INT8 compute is ~256 TOPS, with FP16/INT8 mixed-precision support and optimizations for LLM and Transformer inference; it is one of the mainstay models of domestic edge AI boxes (30TOPS+ models).

Strategic position: In the edge computing box market contested by Rockchip, Ascend, and AMD, the Ascend 510 covers high-density edge deployments with 256 TOPS INT8, and with the CANN + MindSpore ecosystem it is the first choice for self-reliant edge inference.

Core Specifications​

ParameterValue
ArchitectureDa Vinci
ProcessNot disclosed
INT8 Compute256 TOPS
FP16 ComputeNot disclosed
MemoryNot disclosed (varies by board configuration)
Memory TypeNot disclosed
TDP25-70 W (varies by board; requires decent cooling)
Release2024 (edge inference workhorse)

Key Features​

  • 256 TOPS INT8 compute: covers high-density deployment of 30TOPS+ edge models
  • LLM inference optimized: Transformer-friendly
  • CANN + MindSpore ecosystem: domestic self-reliance
  • TaiShan multi-core CPU: strong general-purpose computing

Vendor Information​

ParameterValue
CompanyHuawei Technologies Co., Ltd. (HiSilicon)
Websitehttps://www.hiascend.com
Release2024
ArchitectureDa Vinci

Use Cases​

  • ✅ Edge AI boxes
  • ✅ LLM edge inference
  • ✅ Smart video analytics
  • ✅ Domestic edge deployment