General-purpose computing on graphics processing units facts for kids

General-purpose computing on graphics processing units (GPGPU) is a cool way to make your computer's graphics processing unit (GPU) do more than just show pictures. Normally, a GPU is like a super-fast artist, drawing everything you see on your screen. But with GPGPU, we can teach this artist to solve tricky math problems and handle lots of data, tasks usually done by the central processing unit (CPU), which is like the computer's brain.

Imagine your GPU has tons of tiny workers, all doing simple tasks at the same time. This is called parallel processing. While a CPU has a few very smart workers, a GPU has hundreds or even thousands of less "smart" workers, but they can all work on different parts of a problem at once. This makes GPUs incredibly fast for tasks that involve lots of similar calculations, like processing images or scientific data.

GPGPU started in the early 2000s to make computer graphics look even better. But soon, scientists found that these powerful GPUs could also help with complex calculations for scientific computing.

How GPGPU Started

Using GPUs for general computing became much more common after 2001. This was when GPUs got better at handling complex instructions (called shaders) and working with floating point numbers, which are super important for accurate math.

Problems that involve matrices (like big grids of numbers) or vectors (like directions or forces) were perfect for GPUs. GPUs are built to handle these types of calculations super fast. A big moment for GPGPU was in 2003 when two groups of researchers found that GPUs could solve complex math problems faster than CPUs.

At first, using GPUs for general tasks was a bit tricky. Programmers had to pretend their data was an image or graphic. But then, new programming languages and tools like CUDA (from Nvidia) and OpenCL (which works with many different hardware types) came along. These tools made it much easier for programmers to use the GPU's power without having to think about graphics at all.

A person named Mark Harris, who started GPGPU.org, even came up with the name GPGPU.

How GPGPU Works Today

Many programming tools allow CPUs to send tasks to GPUs. Some popular standards for this kind of parallel computing include OpenCL and Nvidia's CUDA.

OpenCL is like an open standard that anyone can use. It works on many different computer parts, including those from Intel, AMD, Nvidia, and ARM. This means programs written with OpenCL can run on a wide range of devices.

Nvidia created CUDA in 2006. It's a special set of tools that lets programmers use the C programming language to write code for Nvidia's GPUs, starting with the GeForce 8 series.

AMD, another big graphics card maker, launched ROCm in 2016. It's their open-source answer to CUDA, aiming to offer similar features.

Companies like Microsoft also have their own GPU computing tools, such as DirectCompute for DirectX 11. Even mobile devices like Google Android phones and Apple's iOS devices can use their GPUs for general computing tasks, thanks to tools like RenderScript and Metal.

GPU vs. CPU: The Big Difference

Think of it this way:

• A central processing unit (CPU) is like a super-smart manager who can do many different kinds of jobs, but usually one at a time, or a few at a time. It's great for tasks that need a lot of thinking and decision-making, like running your operating system or browsing the web.
• A graphics processing unit (GPU) is like a huge team of specialized workers. Each worker can only do one simple task, but they can all do it at the same time. This makes GPUs perfect for tasks that involve repeating the same calculation many, many times, like drawing all the pixels on your screen or crunching numbers for scientific simulations.

Originally, data only went one way: from the CPU to the GPU to your screen. But with GPGPU, data can now flow both ways. The GPU can process a huge amount of data very quickly and then send the results back to the CPU. This is super helpful for big tasks like analyzing large images or complex scientific data.

For example, a GPU can quickly analyze every tiny dot (pixel) in an image to find edges or calculate average colors. A CPU would have to access the image data much more slowly from its main memory. By moving the data to the GPU's faster memory, GPGPU can speed up these tasks by hundreds of times!

GPGPU is mostly about software and algorithms (sets of instructions), not just special hardware. It's about how we use the powerful parallel abilities of GPUs.

Memory and Power

GPUs have special memory features that help them work fast. They have very large "register files," which are like quick scratchpads for calculations. This helps them switch between tasks very quickly.

However, all this power comes with a cost: GPUs use a lot of electricity, especially when they're working hard. A powerful GPU can use as much power as the rest of your computer combined!

Stream Processing

GPUs are designed to handle "streams" of data. Imagine a stream as a conveyor belt carrying many similar items. A GPU can apply the same operation (a "kernel") to each item on the belt at the same time. This is why they are called "stream processors."

For example, if you want to make an image brighter, a GPU can apply the "increase brightness" kernel to every single pixel in the image simultaneously. This is much faster than a CPU doing it one pixel at a time.

GPGPU works best for problems that have:

• Huge amounts of data.
• Lots of tasks that can be done at the same time (high parallelism).
• Tasks where one calculation doesn't depend too much on the result of another calculation.

What GPGPU is Used For

GPGPU is used in many exciting fields:

• Physics Simulations: Making virtual worlds behave like the real world, like how cloth moves, how fluids flow, or how objects collide in video games.
• Image and Video Processing: Making photos look better, adding special effects to videos, or helping computers "see" and understand images.
• Scientific Computing:

* Weather Forecasting: Running complex models to predict the weather. * Climate Research: Understanding how our planet's climate works. * Molecular Modeling: Simulating how tiny molecules behave, which is important for creating new medicines or materials. * Bioinformatics: Analyzing huge amounts of biological data, like DNA sequences, to understand living things better.

• Machine Learning and AI: Training smart computer programs to recognize patterns, understand speech, or drive cars. This is a huge area for GPGPU!
• Financial Modeling: Doing complex calculations for banking and finance.
• Medical Imaging: Helping doctors analyze X-rays, CT scans, and other medical images.
• Cryptography: Breaking codes or securing information.
• Cryptocurrency Mining: Solving complex math puzzles to create new digital currencies like Bitcoin.
• Antivirus Software: Scanning files quickly to find and stop computer viruses.

GPGPU is a powerful tool that helps computers tackle some of the biggest and most complex problems in the world, making them faster and more efficient!

See also

Fastra II Physics processing unit (PPU) AI accelerator Deep learning processor (DLP)