What is CUDA (NVIDIA)?

The technology used in graphics processors is very different from the technology used in the CPU, for a simple difference: CPU works with exact factors, the GPU does not, it fails a lot.

The risk of you seeing a flaw in a vector is not as important as a wrong calculation of a bank account or a correction of interest, which is why the GPU should never be used for programming; until now.

NVIDIA has been working with cuda processors for a long time, distributed over its GPU, and is able to eliminate these processing errors by applying error correction technologies, reducing its real performance a little, however, it is still much faster than a traditional CPU , even having a physical speed between 300 MHz and 1 GHz. Understand more about.

Understanding its function

The applicability of CUDA technology expands a lot every day, among them, game development companies, who start using the technology to program all actions in games but also for other cases such as disease research, space analysis by NASA, scientific research and even discovering diseases and physical probabilities.

All of these features would probably take days on a typical computer, and with the GPU and CUDA technology, all of this has been greatly reduced.

How can a GPU be "faster" than the CPU?

There is no magic in this matter, the reality is that the GPU's processing chips can only work well in a split way, unlike the CPU that is fully sequential (but with parallel processing, we also see another segment for the CPU tending to use the same as GPUs have been using for a long time).

A CPU contains up to 16 cores for the computer, either physical or virtual (HT technology), dividing the threads and being able to run more programs together. Each core processes its limit, but others are there to work in parallel.

A GPU, in turn, can have 900 and up to 5000 processing cores (GTX 660 or GTX Titan Z, prices vary between R $ 800.00 and R $ 14,000.00). That is, if your programming fails to divide the threads into 1000 pieces to run in parallel, you will not have performance gains when running on a GPU.

In fact, each GPU core is much slower than a CPU, but since it can process a lot of information in parallel, it can deliver a lot of information beyond the ordinary CPU, as long as the programming is aligned to work in parallel, but with the evolution of the CPU, much of the software is already being programmed to take advantage of these parallel computing resources.

When did it come?

The technology was developed by NVIDIA in 2006, benefiting astronomy, biology, chemistry, physics, data mining, manufacturing, finance and other computing-intensive items that benefit from GPU acceleration.

What is NVIDIA Tesla?

According to the NVIDIA website, it is the first processor in the world to have 1 computing teraflop based on various processor colors. It is a solution to deliver more processing power and use electrical resources more efficiently. Tesla is a processor for the biggest computing challenges out there, fast and error free.

OpenACC, what is it?

This is an NVIDIA software development platform, an open standard for programmers to write code in C or Fortran language that allows the compiler to produce software compatible with the parallel processing power on a GPU and at the same time make it compatible to run on a system containing only CPU.

Find more information about CUDA at:

About: https://developer.nvidia.com/cuda-zone

NVIDIA website: https://developer.nvidia.com/cuda-faq

OpenACC: https://developer.nvidia.com/openacc