When it̵7;s time to buy a new computer, upgrade your graphics card or replace your storage, you’ll see one word repeatedly: benchmarks. But how representative are real performance benchmarks?
What is a benchmark?
A benchmark is a test or series of tests designed to drive the performance of your system or component to see what it can do. For graphics cards, this usually means a graphics-heavy scene from a video game or one that may be in a video game. The latter is called a synthetic benchmark, and there are many options, such as Unigine Heaven, 3DMark and PassMark.
For processors, benchmarks are about workload and how fast it can execute instructions. Because there are so many operations that a computer can perform, you will find that different processors work better on one task than others. Some may be better at running productivity software, while others excel at 3D rendering and so on.
There are standard benchmark suites for testing processors, such as PCMark 10, that run your computer through a series of tests. It tests, for example, how your system handles work with spreadsheets, as well as tasks such as photo editing, video calls, physics calculations for games and browsing. Another popular tool for seeing how a CPU handles video playback is CineBench.
CPU benchmarks can also involve specific real tasks, such as compressing a large folder into a ZIP file or loading an application with a large file.
Finally, to test SSDs and hard drives, it is about how fast a device can read and write (save) data to the device. This is usually performed with a benchmark program that performs sequential and random read and write tests.
Sequential means that a large block of data is read or written from adjacent locations on the disk, while random is the opposite. There are also large file tests (about 50 GB) during which the device’s internal cache is stressed (empty cache tends to slow down a device to a scan).
Context is everything
When reviewing benchmarks, you need to keep the context in mind. This includes how one CPU or graphics card performs compared to another, what tests were performed and under what conditions.
Common problems, such as how much RAM a system has, the type of cooling it uses for the CPU and GPU, or how well a case takes in cool air and drives out hot, can all affect performance. Heat is a big deal for computers, as the components block performance the hotter they become as a survival mechanism.
This is good! You do not want components that run themselves until they literally melt or damage sensitive internal parts.
Speaking of heat, even the test room itself can affect performance. A gaming PC works better in a room that stays around 72 degrees Fahrenheit in the summer. It is much harder to keep a computer cool in a warm room.
These are the basic issues to keep in mind for hardware. But each benchmark needs a comparative context to understand the results.
Graphics card comparisons
Generally, players are looking for graphics cards that can beat 60 frames per second. This is the “golden zone”, where games work smoothly and the graphics look really good. All under it, and you encounter stuttering, jumping character movement and low-resolution rendering.
There are two broad considerations to keep in mind when it comes to graphics card performance: resolution and settings. A graphics card may not work well with 4K resolution, but can be an absolute monster at 1080p. This is why it is important to consider the resolution when looking at benchmarks.
When it comes to graphics settings, there are four general automatic presets for video games: Ultra, High, Medium and Low. It can be much more complicated if you manually adjust the settings. However, these four categories are how games are set automatically based on the system’s features. Most reviews use the Ultra setting as a benchmark, unless otherwise stated.
An ideal graphics card can pump out about 70 frames per second or more in 4K with Ultra settings on graphics-intensive AAA games. Cards with this type of performance, however, are generally expensive.
Anyone looking for cards in a budget wants to consider performance versus price. This varies depending on personal preferences and budget.
When reading a review, it also matters which games or synthetic benchmarks were used. Synthetic benchmarks can be useful in comparing one graphics card to another as the test will be consistent from one system to another. The problem is that synthetic benchmarks do not necessarily give a true picture of current video games or what you can expect under real gaming conditions.
Built-in video game benchmarks are not a perfect option either. Many (but not all) games deliver their own benchmarks. However, some of these are not reliable because they are not very active nor do they reflect typical games.
Other benchmarks are better because they use scenes that you are likely to see in the game. Apart from trial and error, there is no real way to know which benchmarks in the game are ideal and which are not.
In addition, a single benchmark is not enough to understand how good a card is. You need several benchmarks to get a complete picture of what kind of performance you can expect.
Let’s look at a real example. Based on the latest reviews, the Nvidia 2080Ti graphics card hits 150-160 frames per second in the game Middle-earth: Shadow of War with 1080p resolution on the Ultra graphics setting. This tells us that the 2080 Ti is an excellent graphics card that works well for this type of game. However, this does not mean that every game will hit these frames.
For example, based on some reviews, the 2080Ti will not exceed 90 FPS on the more intense ones Ghost Recon Wildlands with the same resolution and graphics setting.
If you look at a variety of games and tests, you get a more comprehensive picture of what you can expect from a graphics card before you pick it up in your system.
Processors and benchmark for storage devices
CPU reference numbers are important, but they are most meaningful compared to other processors. Unlike graphics cards, there is no real “golden zone” for CPU performance.
CPUs are workhorses that need to be performed during all types of operations, including games, photo editing, cracking large spreadsheets, or just starting large programs. When looking at processor benchmarks, you want to compare them with what other processors do.
If the processor you want to use for work does not work so well in productivity applications, its game chops do not matter. When it comes to processors, compare them based on what you plan to do with your computer.
The same goes for storage devices. Take a look at the read-write speeds and then compare them with other devices measured in the same review. Also pay attention to large file transfer tests – especially if you move many photos or videos between external storage and your computer.
Finally, remember that benchmarks in reviews usually use stock settings, not overclocking. When you start overclocking a CPU or GPU, you can squeeze out more performance. However, the improvement varies depending on a number of factors, all the way down to the individual build quality of the component you want to overclock.
For example, if you get a processor that works really well in overclocking, it is common to call it “winning the silicon lottery.” This is because it may not have unlocked another CPU with the same model number.
A helpful guide
Comparisons can be a good guide to how computer components work, but the context matters. Compare your components and look at a variety of well-designed tests.
If you always think about how you plan to use your computer, you can get a good sense of what to expect when you hit that valuable new kit in your installation.