Ascend 910C

Headlines · 2

1. 3d
   Huawei, HKUST BalanceRoute boosts decode throughput ~15% in LLM serving tests

   Machine Heart (Jiqizhixin) reports that a team from HKUST led by Zhou Zijie and Huawei 2012 Labs' theory department proposed BalanceRoute for data-parallel load imbalance in LLM serving, where synchronized decoding leaves lighter groups waiting for the busiest. The method models this waiting as an imbalance objective and turns short-term load prediction into online routing (BR-H), with requests pinned to a worker after assignment. In the paper's 96-NPU Ascend 910C experiment running DeepSeek-V3 671B, the authors report about 15% higher decode throughput than the strongest baseline and a 1.20 ms median scheduling time; they say the work was accepted at NeurIPS 2026 and deployed in MathBot-Pie. The figures are author-reported and have not been independently reproduced.

   机器之心 · 🔥 0
2. 5d
   NVIDIA's China AI chip share falls to 34% as Huawei closes gap to 6 points

   According to IDC data cited by 36Kr, NVIDIA's share of China's data center AI chip market fell to 34% in the first half of 2026, with Huawei at 28% — a gap of 6 percentage points versus 31 a year earlier. NVIDIA shipped 678,000 AI accelerators in China from January to June, down 44% year on year, while Huawei shipped 572,000, nearly double its year-earlier total of more than 300,000; NVIDIA's CFO previously said it shipped no Hopper data center products to mainland China in the quarter and took a $4.5 billion inventory charge. Huawei's gains came despite supply constraints, with Ascend 910C yields and high-bandwidth memory supply still capping output, and China's accelerator mix shifted as non-GPU accelerators (1.016 million units, up nearly 80%) exceeded GPUs (991,000) for the first time.

   36氪 人工智能 · 🔥 0

Experience and discussion from the community

Share my Ascend 910C experienceAsk about Ascend 910C