notes.md

notes

• Core gpu design tenets:

  • lot of simple compute units trade simple control for more compute.
  • Explicitly parallel programming model.
  • optimize for throughput not latency

• "cudaMemcpy" is used to copy data between different memory spaces within a CUDA program. It is commonly used to transfer data between the CPU (host) and the GPU (device) memory. The function has the following signature in CUDA C/C++:

cudaError_t cudaMemcpy(void *dst, const void *src, size_t count, cudaMemcpyKind kind);

• "cudaMalloc" is a function for allocating memory on the GPU. It is used to dynamically allocate memory on the GPU device, making it available for use by CUDA kernels and functions.

cudaError_t cudaMalloc(void **devPtr, size_t size);