Talking about GPU

Programmable Graphics Processing Unit (GPU), programmable graphics processing unit, programmable graphics hardware.

The 98 NVIDIA modern GPU was successfully developed using transistors (transistors) for calculations.

Since 03, programmable graphics hardware was formally born, and GPU programming was also announced.

Currently, the latest programmable graphics hardware has the following features:
1. Support vertex programmability and fragment programmability;
2. Support IEEE32 bit floating point arithmetic;
3. Support 4-yuan vector, 4-order matrix calculation;
4. Provide branch instruction, support loop control statement;
5. High-bandwidth memory transfer capability (>27.1GB/S);
6. Support 1D, 2D, 3D texture pixel query and use, and extremely fast;
7. Support Drawing to texture function (Render to Texture, RTT).

Because GPUs have high parallel structures (highly parallel structure), GPUs have a higher efficiency than CPUs in processing graphics data and complex algorithms .

The so-called "parallel computing" means that "multiple data can be used at the same time, multiple data parallel operations and a single data execution time is the same".

Although the GPU uses data parallel processing to greatly speed up the operation, but it is because "any one element calculation does not depend on other similar data", resulting in "need to know the correlation between data" algorithm, on the GPU is difficult to achieve (but on the CPU can be easily implemented), A typical example is the intersection operation of a ray and an irregular object.

In addition, the GPU is weaker than the CPU in the control flow, it can be seen in the graph, the controller in the GPU is less than the CPU, and the main function of the controller is to take instructions, and point out the next instruction in memory location, control and coordination of the various parts of the computer to work methodically.

--excerpt from the spring Snow lowbrow of GPU programming and CG language

Summarize:

-Due to the high parallelism of the GPU, the processing of pixels does not need to be traversed as high-level languages like C + +

-The GPU is not only more efficient at processing graphics data, but also more expressive in dealing with complex algorithms that "do not rely on other types of data"

-GPU provides vertex programming and pixel programming capabilities

Talking about GPU