跳到主要内容Skip to main content

GPUGraphics Processing Unit

Graphics Processing Unit

  • 硬件与部署Hardware & deployment
  • 第五站 · 跑起来Station 5 · Running It
解决什么问题What problem it solves

真正做矩阵运算的芯片。跑大模型靠它,CPU 很慢。The chip that actually does the matrix arithmetic. Large models run on it; a CPU is far too slow.

这是什么What it is

擅长大规模并行计算的处理器,最初为图形渲染设计,现在是大模型训练与推理的主力。

A processor built for large-scale parallel computation, originally designed for graphics rendering and now the workhorse of model training and inference.

为什么需要知道Why it matters

决定你「能不能跑、跑多快」。同一张卡,显存大小比算力型号更卡脖子。It decides whether a model runs and how fast. On the same card, VRAM capacity matters more than the compute tier.

别名Chinese aliases

显卡 · 图形处理器

出处Sources
  • NVIDIA CUDA 工具包文档CUDA Toolkit documentation ↗
  • CUDA C++ 编程指南CUDA C++ Programming Guide ↗