All computer software runs on hardware, so what exactly does it mean for software to use "hardware acceleration" if the hardware is already doing the job? As it turns out, there are ways to use computer hardware faster. It's much faster.
What is hardware acceleration
Hardware acceleration is a method of computation in which calculations are performed much faster using specialized hardware than general-purpose hardware. We'll explain what that means in a minute, but overall, it's much faster if you're using specialized hardware for some common tasks. GPUs are a good example of this. Another historical name for GPUs is "3D accelerators", although today they certainly do more than just 3D graphics.
The CPU in your computer can do the same math as the GPU, only the GPU can do it faster and fast enough to make complex real-time 3D graphics possible. Other examples of hardware acceleration in typical computers and mobile devices are video encoding and decoding. Most modern computers and mobile phones have specialized chips inside them that can process video quickly while using much less power than a full CPU to accomplish the same task.
Large processors like CPUs and GPUs have specialized sections in their own structure dedicated to special work. For example, NVIDIA RTX graphics cards have special areas of the GPU that are used to accelerate ray tracing or AI-related tasks.
Why hardware acceleration is faster
Hardware acceleration is specialization. Your CPU is a jack of all trades, but it's not a panacea. It requires components in the design to perform any computation that can be represented in binary code, but this is inherently inefficient. As a simple example, if you have to build a machine that can add, subtract, multiply, and divide numbers, it is unlikely to be as fast as a machine whose sole purpose is to add numbers.
What are the disadvantages of hardware acceleration
There are some trade-offs to be made with hardware acceleration. The biggest problem is the loss of flexibility. When you have the hardware to do one thing, or a small set of related things, you give up the ability to do something else. That's why we always need a CPU of one kind or another, no matter how many dedicated coprocessors you have.
Another significant disadvantage of hardware acceleration is that you're locked into a specific process. For example, cryptocurrency miners typically use ASICs (application-specific integrated circuits), which are fast and efficient when mining specific blockchains. However, they can't mine on other blockchains that use different "math", so if this form of cryptocurrency becomes unprofitable, the hardware becomes useless. Your phone can speed up video decoding for specific video codecs, but not with newer codecs. You need a new phone!
Similarly, the game Alan Wake 2 uses a graphics hardware component called a mesh shader. If you don't have a graphics card that provides accelerated hardware support for the mesh shader feature, the GPU must switch to more general-purpose hardware for computing, resulting in huge performance issues.
For those designing microprocessors, there is also an important issue with specialized hardware and limited real estate inside the chip. For example, the ray tracing acceleration hardware in an NVIDIA RTC graphics card might be a general-purpose GPU processor. This means that you'll forgo non-ray tracing performance in favor of the feature running faster.
When to turn on hardware acceleration
The simple answer is that hardware acceleration should be turned on as long as it is available. In most cases, the impact on performance is profound compared to using generic processing. For example, hardware-accelerated video exports can have a real financial impact for content creators, as time is money.
However, hardware acceleration isn't always the best solution:
Sometimes hardware acceleration is the cause of application instability. For example, the hardware acceleration capabilities of web browsers often cause problems. That's why turning off your browser's hardware acceleration is a common initial troubleshooting step.
Hardware acceleration can sometimes be faster, but produce lower quality results. So, if precision is a priority, you may want to do things the slow way.
However, if there are no quality or stability issues, then what are you waiting for!