Back    Zoom +    Zoom -
Tencent Hunyuan AI Infra Launches HPC-Ops, Boosting Inference Throughput by 30%
Recommend
10
Positive
22
Negative
11
TENCENT (00700.HK)'s LLM Tencent Hunyuan AI Infra team announced the launch of HPC-Ops, an open-source production-grade high-performance LLM inference core operator library.

Based on production environment challenges, it is built from scratch using CUDA and CuTe. By abstracting the engineering architecture, deeply adapting the microarchitecture, achieving extreme optimization at the instruction level and others, the team lowers the threshold to developing low-level operators, pushing core operator performance to approach hardware peak, and enabling breakthrough in performance.

Related NewsCiti Expects Limited Negative Impact of Google's Project Genie on CN Gaming Industry, Reiterates Buy on TENCENT
In real-world scenarios, the Hunyuan model's inference QPM increased by 30% based on HPC-Ops, while the DeepSeek model's QPM rose by 17%.
AASTOCKS Financial News
Website: www.aastocks.com