What is GPGPU? Technology Leveraging High Parallel Processing Performance Beyond Graphics

In recent years, “GPGPU” has been attracting attention in fields such as video editing and AI development.

GPGPU is a technology that utilizes GPUs, which were originally designed only for graphics processing, for general-purpose computation. Thanks to its overwhelming processing speed, it can greatly improve work efficiency.

This article will explain the basics of what GPGPU is, and introduce easy-to-understand examples of its use in video editing, AI training, scientific computing, video production, mining, and more.

In addition, the article will explain the performance differences between CPUs and GPUs, as well as the actual effects based on real experiences.

The article also explains basic knowledge such as how to read graphics card manufacturers and model numbers, performance indicators, and how to choose a GPU from the perspective of performance and compatibility.

≫ Related article: How to Choose a Graphics Card for a Custom PC [Performance / Features / Compatibility]

あわせて読みたい

PC Parts Estimation & Compatibility Check Tool

Automatically checks each component you need when selecting parts for a custom-built PC—perfect for beginners.

What is GPGPU?

This section explains the basic role of GPUs and their new applications.

GPUs are Responsible for Graphics Processing

A GPU (Graphics Processing Unit) is a specialized processor mainly responsible for calculations related to graphics, such as effect processing during image editing, rendering in 3D games, and outputting images to monitors.

As a PC component, it is installed on a “graphics card“, and the term GPU refers to the actual processing chip on the graphics card.

For gamers, it is an almost essential part, so it may be very familiar.

Strictly speaking, there are differences, but since their uses and structures are closely related, it is generally fine to think of “GPU = graphics card”.

GPGPU: Using GPUs for More Than Graphics

GPGPU (General Purpose computing on Graphics Processing Units) is a technology that utilizes GPUs, which were originally designed for graphics processing, for general-purpose numerical calculations such as scientific computing and machine learning.

GPUs were originally specialized for graphics processing, such as image processing and video output, and were rarely used for other purposes.

However, around 2006, NVIDIA announced CUDA, a platform that allows general-purpose computation directly on GPUs. Since then, the high parallel processing capability of GPUs has attracted attention, and they have started to be used for purposes other than graphics processing.

GPUs have a structure with many processing units in parallel, and are excellent at processing large amounts of data at the same time.

Because of this, large-scale data processing and repetitive calculations that used to take a long time on CPUs can now be executed much faster and more efficiently.

Today, GPGPU is used in various fields such as machine learning, weather simulation, genome analysis, and financial modeling. In particular, its capabilities are essential for deep learning training.

In this way, GPUs have gone beyond being just processors for graphics, and now also play a role as processors for general-purpose computation.

Examples of GPGPU Usage

This section introduces specific examples of how GPUs are used for processing other than graphics.

Video Editing Encoding

Encoding is one of the heaviest tasks in video editing.

Encoding refers to the process of outputting and saving edited video data as a single video file.

This process requires a large amount of computation to compress and save video data, and processing time tends to be long, especially for high-resolution videos.

By using GPGPU technology, the parallel processing power of the GPU can be used for encoding tasks.

Many video editing software programs support GPGPU, and by assigning part of the encoding process to the GPU, encoding time can be greatly reduced compared to using only the CPU.

GPGPU is especially effective for high-resolution videos such as 4K and 8K, or projects that use many complex effects.

With improved processing speed, creators can produce high-quality videos in less time, greatly improving work efficiency.

AI Training (Deep Learning, Machine Learning, etc.)

In the field of AI, using GPGPU greatly improves the efficiency of training processes.

Especially in deep learning, training neural networks with large amounts of data requires very large computational resources.

This training can be done on a CPU, but using a GPU can make it 20 to 50 times faster in some cases.

However, not all AI algorithms can use GPUs.

For example, GPUs are actively used in deep learning, but some methods, such as genetic algorithms, may not be able to use GPGPU.

Also, whether a GPU can be used depends on “which library is used” and “which model is built” when programming.

That said, AI training is the most demanding and time-consuming part of AI development.

With a GPU, tasks that used to take a week can sometimes be completed in a few hours, making the development and testing cycle much faster.

In addition, shorter training times make it easier to increase the amount of data or the number of model parameters to improve accuracy.

For those who want to learn AI or actually try running AI, choosing a high-end GPU is very valuable.

Scientific Computing and Simulation

GPGPU is also used in scientific computing and simulation.

Large-scale data processing and complex numerical calculations that used to take a long time on CPUs can now be executed much faster by utilizing the high parallel processing power of GPGPU.

For example, in simulations of physical phenomena such as weather forecasting or fluid dynamics, where many elements interact with each other, it is necessary to process a large number of calculations quickly.

GPGPU excels at processing these calculations in parallel, contributing to both processing speed and simulation accuracy.

In research fields such as molecular dynamics and astrophysics, using GPUs makes it possible to build higher-resolution and more realistic models, enabling detailed analysis that cannot be obtained through experiments.

In this way, GPGPU is widely used as a means to eliminate bottlenecks in scientific computing and dramatically speed up research and technology development.

Image Generation and 3DCG Rendering

Fields such as image generation and 3DCG rendering are originally classified as graphics processing, which GPUs excel at.

However, in recent years, GPGPU technologies such as CUDA and OpenCL have also been actively used behind the scenes in graphics processing.

For example, 3D model rendering requires processing many calculations at the same time, such as light reflection, shadows, and texture expression.

By using GPGPU for these processes, faster and more realistic rendering is possible, greatly reducing production time.

Furthermore, in the field of image generation using deep learning, GPGPU is a core technology.

Image generation AI requires large-scale matrix calculations and parameter updates using neural networks, so fast computation with GPGPU is essential.

In this way, in the fields of image generation and 3DCG rendering, the combination of traditional GPU graphics processing and GPGPU general-purpose computation enables higher-quality and more efficient visual expression.

Game Recording and Streaming Encoding

In game recording and streaming, it is necessary to encode (compress) gameplay footage in real time and save or stream it as high-quality video.

This encoding process originally placed a heavy load on the CPU, but now, by using GPGPU technology, the GPU can handle part of the process.

With GPGPU, the GPU can process both game rendering and video encoding for recording or streaming in parallel.

Representative technologies include NVIDIA’s NVENC and AMD’s AMF, which are widely used in streaming software such as OBS.

These hardware encoders are a type of GPGPU and can deliver high-quality streaming and recording while maintaining game frame rates and placing almost no load on the CPU.

GPGPU encoding support is especially effective when handling 4K or high frame rate video, making it possible to enjoy both smooth gameplay and streaming.

Real-Time Video Processing and Effects

GPGPU is also being used more in real-time video processing and visual effects.

It is especially effective in interactive video production and live visual performances that require massive calculations to be processed instantly.

By utilizing the parallel computing power of GPUs, real-time processing that was difficult with CPUs is now possible.

For example, in processing complex visual effects such as light reflection, shadow generation, and simulations of natural phenomena like fluids or smoke in real time, GPGPU plays a central role.

Recently, tools such as TouchDesigner and Unity, which can generate, manipulate, and display visuals in real time, and VJ (Visual Jockey) performances that change visuals in response to music or MIDI controller operations, have made GPU-based processing common.

As a result, GPGPU technology has become essential in live performances and interactive art.

In this way, GPGPU is a technology that enhances the expressiveness and responsiveness of video processing, bringing great potential to the entertainment and creative industries.

Physics Engines

GPGPU greatly contributes to real-time simulation in physics engines.

A physics engine is software that reproduces real-world physical phenomena such as gravity, collisions, friction, and the movement of rigid bodies or fluids on a computer.

It is used in games, movie CG, and simulation tools to realistically express objects falling or bouncing.

These processes require massive calculations, but by using GPGPU, physics calculations can be accelerated using the parallel processing power of the GPU.

For example, simulating hundreds or thousands of objects at the same time or particle-based fluid simulations, which would take a long time on a CPU, can be calculated in real time with GPGPU.

This processing power directly leads to improved game experiences that respond instantly to player actions and greater immersion in video works.

Representative technologies include NVIDIA’s PhysX and the physics simulation features integrated into game engines like Unity and Unreal Engine, all of which are designed for fast computation with GPGPU.

Cryptocurrency Mining

GPGPU is also used in cryptocurrency mining.

Cryptocurrency mining refers to the process of continuously calculating hash values that meet certain conditions to generate new blocks on the blockchain.

In blockchain, transactions are grouped into “blocks” and recorded in a chain. To connect a new block correctly, a special “answer” called a hash value must be found.

It is like a puzzle where “this block needs a hash with these conditions! But no one knows the answer, so all machines around the world compete to find it by calculating as much as possible!”

This process requires a huge number of calculations (hash calculations) to be repeated in a short time, and using only the CPU is inefficient and unrealistic.

GPGPU enables efficient processing of the calculations needed for mining.

GPUs can operate thousands of processing units at the same time, making them very suitable for simple but repetitive tasks like hash calculations.

As a result, compared to CPU-only environments, GPGPU achieves overwhelming processing speed and mining efficiency.

Performance Differences and Processing Speed: CPU vs GPU

This section explains the differences in performance and processing speed between CPUs and GPUs.

Understanding Performance Differences from Specifications

First, let’s look at the rough performance differences between CPUs and GPUs.

When comparing performance, “clock speed” and “number of cores” are important, but CPUs and GPUs have completely different design philosophies (architectures).

Therefore, it is not possible to simply compare which is better just by looking at the numbers.

Here, the specifications are compared to give an idea of how structurally different CPUs and GPUs are.

For example, comparing high-end CPUs and GPUs:

Let’s also compare mid-range CPUs and GPUs.

Comparing only clock speeds, GPUs are about half that of CPUs.

However, in terms of core count, GPUs have an overwhelmingly larger number, with differences of 220 or even 682 times compared to CPUs.

By having this huge number of cores process tasks in parallel, GPUs can deliver much higher performance for certain tasks, even if their clock speed is lower.

Of course, the optimal processor depends on the accuracy and type of computation, so it is difficult to simply judge superiority by clock speed and core count alone.

Still, this gives an idea of the scale difference between CPU and GPU architectures.

Video Editing Encoding Time Benchmark

Here is a comparison of encoding times in video editing using GPGPU technology.

The equipment is a bit old, but it was high-end at the time.

The video editing software used is Filmora.

The test environment for the source video and settings is as follows:

As a result, encoding time was reduced by 3 minutes and 57 seconds (277 seconds), making it about 60% faster.

Depending on the CPU and GPU used, when comparing high-end CPUs and GPUs, encoding time can be less than half.

AI Training Time Benchmark

Next is a comparison of AI training times on CPU and GPU.

At that time, one of the AI algorithms called deep learning was used, and AI training was performed on tens of millions of data points.

The overall flow of the program was roughly: retrieving data from the database, AI training, and checking and saving the results.

Of these, GPGPU can be used for AI training, so the time for this part was compared between CPU and GPU.

As a result, it was about 20 to 30 times faster.

This result is based on my experience working on AI-related tasks before starting this blog, comparing CPU and GPU execution times from logs output during training.

Although I do not have detailed benchmark data, I remember being very impressed by how much faster processing became when using a GPU.

Summary: For Faster Video Editing Encoding and AI Study, NVIDIA GPUs are Recommended!

This article explained the basics of GPGPU and in which fields it is used.

Here is a summary of the key points.

GPGPU is a technology that utilizes the parallel processing power of GPUs, originally used for graphics processing, for general-purpose computation such as scientific computing, AI training, and video encoding.

Compared to CPUs, GPUs can process large amounts of data in parallel, so a dramatic speedup can be expected in high-resolution video editing and AI model training.

In fact, video encoding time can be less than half, and AI training can be 20 to 30 times faster.

For these purposes, NVIDIA GPUs are the easiest to use for GPGPU technology.

NVIDIA provides unique platforms such as CUDA, and there are many compatible software and libraries.

For those who want to “edit videos more efficiently” or “learn and implement AI”, NVIDIA GPUs are a great choice.

If you are thinking about building a PC or assembling one yourself, be sure to consider GPGPU when choosing a GPU.

The article also explains basic knowledge such as how to read graphics card manufacturers and model numbers, performance indicators, and how to choose a GPU from the perspective of performance and compatibility.

≫ Related article: How to Choose a Graphics Card for a Custom PC [Performance / Features / Compatibility]

あわせて読みたい

PC Parts Estimation & Compatibility Check Tool

Automatically checks each component you need when selecting parts for a custom-built PC—perfect for beginners.