ما هو CUDA فعليًا

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 NVIDIA's platform for general-purpose GPU computing. It lets you run your own programs on the GPU, not just graphics.

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.

A function you mark to run on the GPU is called a kernel. Thousands of threads will each run that one function in parallel.

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.

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.

CUDA gives you a friendly runtime API: functions like cudaMalloc and cudaMemcpy that manage GPU memory and move data around for you.

You build CUDA code with nvcc. It splits your file, sends host code to your normal compiler and device code to the GPU compiler.

CUDA runs only on NVIDIA GPUs. That tight coupling is why it is so well tuned, but it also means it is vendor specific.

CUDA ships tuned libraries like cuBLAS for matrices and cuDNN for deep learning, so you rarely start a hard problem from scratch.

Mature tools, fast libraries, and deep ecosystem support made CUDA the default way to do GPU computing in science and AI today.

Let us confirm what CUDA really 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. 👍