[Hardware]-What processes images faster, CPU or graphics card?

[ArInA-pAn]

 

 

Every day we take care of mani[CENSORED]ting images with our PC, not in vain our monitor sends us a minimum of 60 of them per second. But, which component processes them faster? The GPU that is specialized in it or the CPU with its great speed. Many of you already know the answer, but we explain why.

 

Image mani[CENSORED]tion is one of the professional disciplines most related to the PC today and for decades. From photo mani[CENSORED]tion to computer art creation, through prepress and editorial design. Day after day millions of images are mani[CENSORED]ted by the hardware of our PC.

 

What is an image

 

We all know what an image is, but the question is more directed at what an image is when we talk about computing, and that is that it is nothing more than a mesh of pixels, each pixel storing the information of the RGB color components.

When it comes to mani[CENSORED]ting an image, what the processor, be it a CPU or a GPU, does is mani[CENSORED]te all those pixels individually and in parallel, mani[CENSORED]ting each and every one of the color components that make up each of the pixels of the picture.

 

 

CPUs are designed to work seriously and not in parallel, this is due to how the Von Neumann architecture on which all CPUs are based works, this means that when mani[CENSORED]ting images they usually do it taking pixel by pixel, and in certain cases even component by component, individually.

 

For reasons of space, there are not several ALUs with different precisions, but to operate with a lower precision data, higher precision ALUs are used. This means that if the CPU has to manage a pixel with 3 or 4 components, RGB or RGBA, then it will manage all one by one each of these when calculating them, which will require many more cycles.

 

Shifting GPUs are meant for graphics processing and their pixel mani[CENSORED]tion instructions are meant to mani[CENSORED]te all 3 or 4 components of a pixel. In addition, the way in which they use memory means that despite the fact that the access is serial as a CPU, it can process several pixels at the same time that it will process making use of its high level of parallelism.

 

Because storage space is important, compressed image formats were developed in which the image information is not given as is, but is encrypted to occupy less space and requires decryption to obtain it.

 

The graphic processors in the beginning were a cluster of fixed function units that could not be programmed, this led to the fact that when using certain image formats, the CPU had to do the decryption, which caused the image to already decoded was in the memory space of the CPU.

 

With the arrival of shader units in GPUs, the hardware of graphics cards began to be used to decode images, but it was a momentary solution, since although it frees the CPU from the task, it was not entirely elegant, that is why Fixed-function units and / or accelerators began to be used to decrypt certain image formats much faster and without affecting CPU and GPU performance.