You are testing your PC performance wrong—here is the right way

There are two ways to measure the performance of PC components: synthetic benchmarks like 3DMark or PCMark, and real-world benchmarks that test performance in games and various software using real-life scenarios. The latter's much better, here's why.

Synthetic benchmarks aren't very good at reflecting real-world performance

Synthetic benchmarks help you gain insight into the ballpark performance of different PC components, see how they stack up against other models, and check whether your CPU, GPU, or some other component works as intended. They're also a great way for enthusiasts to compete, especially if a benchmark includes leaderboards.

Putting your GPU through the 3DMark Steel Nomad or Time Spy benchmarks and then comparing the result against scores other people achieved with the same GPU can inform you whether your graphics card works as intended. If you get a score that's way lower than expected, something's likely wrong, and it's time to troubleshoot your PC .

The same goes for other components. When I tested the ROG Xbox Ally X's CPU in Cinebench, my single-core score was much lower than what other people had reported. So I started troubleshooting and discovered that the Windows power mode was set to "Balanced," which limited the CPU's maximum frequency and thus hampered its single-core performance. After I switched to the "High Performance" power mode, my Cinebench single-core score shot up to a level expected from the Z2 Extreme APU.

What synthetic benchmarks aren't good at is accurately reflecting real-world performance across different games and apps. A great example of this is the MSI Claw 8 PC gaming handheld. In my ASUS ROG Xbox Ally X review , the Intel Arc 140V integrated graphics found in the MSI Claw 8 destroyed other handhelds in the graphics portion of the Time Spy benchmark, as well as in the Steel Nomad benchmark, which focuses on GPU performance.

But once you scroll down to the game benchmarks, you can see that the MSI Claw 8 usually ended up behind or only slightly ahead of the ASUS ROG Xbox Ally X instead of mopping the floor with it, like you'd expect to happen after seeing those 3DMark numbers.

The reason for this is that many games don't perform the same on equivalent GPUs from different vendors—NVIDIA, AMD, or Intel—either because of drivers, the engine a specific game uses, or simply because the game itself is better optimized for certain GPUs. Synthetic GPU benchmarks, however, usually measure raw performance, the "best case scenario" if you will, where drivers and game optimizations play a minimal role.

For instance, newer Call of Duty games run faster on AMD graphics cards than on their NVIDIA and Intel equivalents because they're optimized for AMD GPUs. On the other hand, the majority of games with ray tracing will run faster on NVIDIA cards than their AMD and Intel equivalents once you enable ray tracing effects, because NVIDIA GPUs have the best ray tracing support, both hardware and software-wise.

NVIDIA GPUs are also the best choice for professional work, such as GPU rendering or AI workloads, because their proprietary CUDA GPU compute platform is way ahead of what AMD and Intel GPUs offer.

The same goes for CPUs. Apps either favor single-core or multicore performance, with some being better optimized for AMD processors and others for Intel. Similarly, most games favor X3D CPUs from AMD because they offer excellent single-core performance and pack much more L3 cache memory than your average CPU, and games like two things: fast single-core performance and lots of L3 cache.

The only difference compared to GPU synthetic benchmarks is that some CPU synthetic benchmarks are actually quite good at gauging real-world performance for certain tasks. Cinebench and Blender benchmarks, for instance, are excellent for evaluating how a specific CPU model performs in rendering workloads because both benchmarks include real rendering tasks.

Focus on real-world benchmarks to learn about the actual performance of a particular CPU or GPU

Instead of synthetic benchmark scores, you should focus on real-world benchmarks to find out the level of performance of a specific CPU or GPU. For gaming performance, these include either built-in or custom scene in-game benchmarks. When it comes to productivity, focus on benchmarks that test real-world workloads, such as the time required to render a scene in Blender or export a set number of photos in Adobe Lightroom .

When in the market for a new CPU or GPU, you should focus on the tasks you're actually interested in using it for. If you're buying a new GPU for AI workloads, don't check gaming benchmarks; look up how the graphics card you're interested in handles AI image generation or whichever AI-related workloads you plan to use it for.

Conversely, if you're in the market for a new gaming GPU, check how it performs in games you're interested in playing. For example, if you plan to play AAA games with every bell and whistle enabled, focus on gaming performance with ray tracing, since most new AAA games support ray tracing effects.

It's a similar story with CPUs. If you're a PC gamer, you should focus on gaming benchmarks. But if you're looking for a new workstation CPU, check its performance in the apps you use in your daily workflow.

Don't limit yourself to just a couple of real-world benchmarks—the more you check, the better

Now, only checking a single real-world GPU or CPU benchmark is far from ideal. Different reviewers will end up with different scores and frame rates depending on which CPU and RAM they use, whether they opted for a built-in gaming benchmark or used a custom in-game scene for testing, the scenario they used for real-world CPU and GPU productivity benchmarks, and so on.

More data is always better, so try looking up multiple sources to get a better idea of a particular CPU or GPU's performance. Luckily, there are a number of trustworthy sources of CPU and GPU reviews out there for you to peruse.

If you prefer video form, I recommend Gamer's Nexus and Hardware Unboxed . Daniel Owen also has excellent real-world gaming GPU tests. I can also recommend Digital Foundry for CPU and GPU reviews that focus on gaming performance.

When it comes to written form, Gamers Nexus publishes its reviews on its website some time after releasing them on YouTube, and TechSpot publishes Hardware Unboxed's tests in written form. Then you've got PCWorld , KitGuru , and Tom's Hardware , all of which offer pretty good PC component reviews. Lastly, Puget Systems is an excellent resource for information on CPU and GPU performance across various productivity tasks.

While I contribute to the website and am thus biased, I also have to mention TechPowerUp because the outlet publishes in-depth CPU, GPU, and other PC component reviews, with CPU reviews featuring a bunch of real-world tests. TechPowerUp, Daniel Owen, and Hardware Unboxed also publish GPU benchmarks for newly released AAA games, where they benchmark them across a wide range of contemporary GPUs.

---

It's not that synthetic benchmarks are inherently bad; they just aren't the best source of information about real-world GPU, CPU, or other PC component performance.

They're useful, for example, when you want to verify that your CPU or GPU is performing as expected. But if you want insight into real-world performance, focus on gaming and productivity benchmarks that use real-life workloads.