https://gpu.itbrief.co/posts/tencent-optimizes-ai-efficiency-amid-gpu-constraints/