CUDA 到底是什么
NVIDIA 用于在 GPU 上运行 C++ 的平台
CUDA 到底是什么 是 CoddyKit 上的免费 CUDA Academy 课时。 这是第 3 节课，共 4 节。 你可以在下方免费阅读本课时的完整内容 — 然后在浏览器中使用内置代码编辑器和全天候 AI 导师进行实践。 这是 CUDA Academy 学习路径的一部分，你的进度在网页和 CoddyKit 应用中同步。 CUDA Academy 课程共包含 4 节课。
本课时的部分内容尚未翻译，以英文显示。
Putting a Name to the Magic
You have seen why GPUs are fast. CUDA is the tool that lets you actually write code for them in C++. Time to meet it properly. 🚀
CUDA Is a Platform
CUDA is NVIDIA's platform for general-purpose GPU computing. It lets you run your own programs on the GPU, not just graphics.
An Extension of C++
CUDA C++ is just C++ with a few extra keywords and functions. If you know C++, most of CUDA already feels familiar from day one.
Kernels: Code That Runs on the GPU
A function you mark to run on the GPU is called a kernel. Thousands of threads will each run that one function in parallel.
__global__ void add(int *out, int *a, int *b) {
// runs on the GPU
}The __global__ Keyword
That __global__ tag marks a function as a kernel: callable from the CPU but executed on the GPU. It is your gateway onto the device.
Host and Device
In CUDA talk, the CPU side is the host and the GPU side is the device. You will hear these two words constantly, so lock them in now.
The Runtime API
CUDA gives you a friendly runtime API: functions like cudaMalloc and cudaMemcpy that manage GPU memory and move data around for you.
nvcc: The CUDA Compiler
You build CUDA code with nvcc. It splits your file, sends host code to your normal compiler and device code to the GPU compiler.
It Is NVIDIA-Only
CUDA runs only on NVIDIA GPUs. That tight coupling is why it is so well tuned, but it also means it is vendor specific.
A Rich Library Ecosystem
CUDA ships tuned libraries like cuBLAS for matrices and cuDNN for deep learning, so you rarely start a hard problem from scratch.
Why CUDA Took Over
Mature tools, fast libraries, and deep ecosystem support made CUDA the default way to do GPU computing in science and AI today.
Quick Check
Let us confirm what CUDA really is.
Recap: What CUDA Is
CUDA is NVIDIA's platform that extends C++ with kernels, a runtime API, and the nvcc compiler so your code can run on the GPU. 👍
常见问题解答
「CUDA 到底是什么」课时是免费的吗？
是的 — 「CUDA 到底是什么」的完整文本可在网页上免费阅读。要进行交互式练习（内置代码编辑器和全天候 AI 导师）并解锁 CUDA Academy 课程的其余内容，请升级到 CoddyKit PRO。 CUDA Academy 课程共包含 4 节课。
「CUDA 到底是什么」这节课中我会学到什么？
NVIDIA 用于在 GPU 上运行 C++ 的平台 你通过在浏览器中直接运行的动手代码来练习 CUDA Academy，全天候 AI 导师会在你学习这节课的过程中回答你的问题。
学习 CUDA Academy 需要有经验吗？
无需任何先前经验。CoddyKit 上的 CUDA Academy 课程适合初学者到高级学习者，你可以从这里开始或从头开始，按照自己的节奏学习。 这是第 3 节课，共 4 节。
「CUDA 到底是什么」课时需要多长时间？
大多数 CoddyKit 课程大约需要 5–10 分钟。每节课都很精短且互动，所以你能稳步进步，并在网页和应用中从离开的地方继续。
我能在这节 CUDA Academy 课中编写并运行代码吗？
能。每节 CUDA Academy 课都包含内置代码编辑器，你可以在浏览器中直接编写并运行真实代码，并获得即时 AI 反馈 — 无需本地设置。