Local Data Storage News Today : Breaking News, Live Updates & Top Stories | Vimarsana

This article summarizes some lower level aspect of how GPU executes. Although GPU programming is not that complicated when compared to CPU, it also doesn’t match to what hardware is doing exactly. The reason is that we can’t just program GPU without some API, which is an abstraction over its inner workings. Since few years now, we have modern explicit APIs like DirectX 12 or Vulkan, which shrunken the gap to what is happening with hardware. Yet there still are few low-level bits (pun intended) that are worth explaining.

Nvidia gpusSingle program multiple threadsMemory bankRegister bankSpecialized procesing unitShared memoryStreaming multiprocessorCompute unitWorkgroup processorExecution unitTensor coresBounding volume hierarchyHyper threadingSpecial purposeFused multiplyLocal data storage