Understanding CPU benchmark testing

Benchmarks do not always accurately reflect real-world performance.

Upgrading to a new CPU is both exciting and frustrating. Users need to check the specifications and benchmarks of the CPUs they are interested in to see how well they perform. This article will specifically introduce how to read CPU benchmarks.

Synthetic Benchmarks vs. Real-World Benchmarks

Synthetic benchmarks are standardized performance testing tools that push the CPU to its limits in a controlled environment. The most popular ones are Cinebench and PCMark, which users can run on their computers to test CPU performance, but it may be easier to look at the results others have obtained.

Synthetic benchmarks are a good way to test the raw performance of a CPU because they push each core to its limit, but the scores do not strictly indicate better performance, at least not in all scenarios. For example, an older AMD Threadripper might outperform the latest Intel Core i7 in synthetic benchmarks, but the i7 might surpass the Threadripper in most games and even some productivity tools.

Real-world benchmarks, on the other hand, are the complete opposite. They test performance using tools and video games. Some programs and games have built-in benchmarks, making testing easier, but if not, hardware testers will create their own tests and use these tests on all CPUs to provide accurate results.In general, actual benchmark tests are more representative of CPU performance, especially if users need the CPU to perform specific tasks. Some programs and games prefer more cores and threads, even if they are slower, while other programs only use a few cores, in which case users hope for faster clock speeds. For example, if game performance is a priority, one can look at video game benchmark tests to see which CPU performs best in the user's favorite games.

Advertisement

Single-core performance vs. multi-core performance

Single-core performance and multi-core performance are terms used in synthetic benchmark tests. As the name suggests, single-core performance indicates how fast a single CPU core is, while multi-core tests the performance when all cores run together.

Single-core performance is important for tasks that run on a single thread, such as most simple programs like text editors, spreadsheets, and game launchers. Multi-core performance is more important for high-demand tasks that utilize multiple cores, such as video and image editing, 3D rendering, and gaming. However, since these programs mostly cannot use all cores at the same time, these numbers are not the best way to compare performance.

Productivity benchmark tests

If a good CPU is needed for work, one should look at productivity benchmark tests. More specifically, users should check the benchmark tests for the tasks and programs they will use. For example, any tasks that are mainly focused on machine learning mainly depend on the GPU, so there is no need to overspend on the CPU. For tasks such as video editing, compilation, and 3D modeling, a good CPU is needed. However, there are always exceptions in specific applications. For example, some video editors provide GPU acceleration to speed up certain tasks, and there are some 3D modeling software packages that use CPU rendering, so they will utilize as many CPU cores as possible.

Here, optimization plays an important role, as some programs may be better optimized for AMD or Intel. For example, many Adobe tools, such as Premiere Pro, seem to favor Intel. Users should check the specific models they are interested in for the most relevant scores.Resolution

Has anyone noticed that most CPU benchmarks focus on the performance of games at 1080p resolution? This is because at high resolutions, performance is more limited by the GPU rather than the CPU, so the performance gap between different CPUs becomes smaller.

But this does not mean that CPU performance is not important for high-resolution gaming. On the contrary, users still need a powerful CPU to keep up with the GPU to maintain smooth FPS.

Although average FPS may not reflect this, a low-end CPU can lead to noticeable stuttering, especially in open-world games. In addition, if the user upgrades their GPU later, the performance gap between slower and faster CPUs will increase significantly. The key here is that if you focus strictly on CPU performance, then benchmarking with a high-end GPU at 1080p resolution is the most important.

Average FPS, 1% Low FPS, and Frame Time

All other factors being equal, the number of frames per second (FPS) achieved by the CPU is one of the best performance indicators. A higher FPS equals better performance. The average FPS tells users how many FPS they can usually get during normal gameplay with the same PC configuration as the hardware testers.

In addition to the average FPS, a good benchmark will also include the 1% and 0.1% minimum values, representing the lowest FPS in 1% and 0.1% of the time, respectively. Although they may seem like random performance data, they are actually extremely important indicators of overall performance. If the 1% and 0.1% minimum values are significantly lower than the average FPS, it indicates the presence of stuttering, and these numbers should be as high as possible.

It is worth noting that optimization plays an important role in FPS. Some games tend to have a CPU with a high core count for strong multi-threaded performance, such as Cyberpunk 2077. Current games usually reach peak performance at six cores. Simpler games, such as most esports games, favor extremely fast single-threaded performance. In addition, all games benefit from more CPU cache.

Some hardware testers also include "frame time" in their benchmarks, which is the time interval between two frames divided by the screen's refresh rate. It is not an average but is displayed in real-time, and a lower frame time is better because there is no delay between frames. Frame time itself is not a performance indicator, but an exceptionally high frame time may indicate a problem with the configuration, drivers, or the game itself.Like most things in life, a margin of error should always be taken into account. The CPU performance seen in a specific benchmark test may vary by a few percentage points, so there is no need to be overly concerned.

Clock Speed

In isolation, clock speed does not tell a user how well a CPU performs, but it is still a benchmarking metric worth paying attention to. Specifically, most consumer CPUs nowadays have a maximum turbo clock frequency. The longer a CPU can maintain its maximum clock frequency, the better its performance will be. Cooling is usually the limiting factor, but if the benchmark test uses a high-end CPU cooler, the CPU should be able to maintain a clock frequency close to its maximum. If it cannot, then the maximum turbo clock frequency in the specifications may be somewhat misleading, or there may be other factors slowing down the CPU's speed.

CPU Utilization

CPU utilization refers to the degree to which the CPU is being used at any given moment, expressed as a percentage. A 60% CPU utilization means that the game is using 60% of the CPU's maximum resources. It is normal to see lower CPU utilization on CPUs with many cores and near 100% utilization on CPUs with four or fewer cores. If a CPU is close to its maximum utilization, it can hardly do anything else besides running the game, while a CPU with only 60% utilization may still be able to run a few Chrome tabs and applications in the background without significantly affecting performance.

This metric is worth noting when two CPUs achieve similar FPS in most games. If CPU #1 consistently maintains a utilization of 95%, while CPU #2 never exceeds 60%, then CPU #2 is undoubtedly the winner, as the unused power can be used to run background applications.

The temperature at which a CPU operates not only affects performance; overheating can actually shorten the CPU's lifespan. Although the "normal" temperature range for different CPUs under full load varies, no CPU should always operate at its maximum operating temperature unless there is a good CPU cooler.

However, some CPUs are inevitably going to run very hot, so pay attention to the CPU temperature in the benchmark test. Like CPU utilization, if a CPU achieves the same performance at a lower temperature than similar models, it may be a better CPU. Alternatively, a powerful AIO cooler can be used to deal with a CPU that generates a lot of heat.Power consumption is one of the most interesting metrics, as it can be significantly different from the TDP (Thermal Design Power) listed on the specifications. This is because TDP is a power rating related to heat dissipation; the actual maximum power consumption is often much higher than the TDP.

Benchmarks will inform users about the CPU's performance in real-world scenarios, such as power consumption in gaming and productivity tools. If one attempts to choose a power supply unit (PSU) based on its power rating reasonably, rather than simply purchasing an overly large model, they may end up with an underpowered PSU if they do not consider the actual power consumption of the CPU.

Purchasing a new CPU becomes much easier if one knows which benchmark metrics to focus on. Thanks to the crash course in this article, users now have the knowledge needed to compare CPU performance. They no longer just prioritize synthetic benchmarks and the highest numbers in games, but understand the importance of those other metrics that are less often mentioned.