Highlight recent changes
The choice of processor determines the performance and efficiency of a computer. But what is the best GPU for gaming? Which CPU should you install for socket 1200, AM4/AM5 or 1700? Which CPU has the best value for money and which CPU goes best with an RTX 4080 or RX 7900 XT? In our benchmark list of the best, we summarize the most important purchase criteria: the performance. We consider not only games, but also and applications. We give the index once as a whole and then again divided into games and applications. This allows you to choose the list that best suits your needs. All values are normalized, which means that all benchmarks are evaluated equally.
CPU tests 2023: This is new
Update from 03/03/2023: With the Alder Lake refresh and the AMD Zen 4 non-X Ryzens, we have a lot of newcomers in the CPU index. The first Zen 4 3D is also included. Furthermore, new benchmarks were introduced, so that the numbers of today cannot be compared with those of the past.
Every year, not only PC games, 3D graphics and hardware reinvent themselves, but also our demands on our benchmarks. Loyal readers know that we only made extensive optimizations and changes to the test course for processors in issue 04/2021. Many of these insights remain; we are concentrating mainly on gaming performance again this year, but are raising the bar with new benchmarks and changed test conditions.
You can find the complete list of games and programs used in the dedicated article, the CPU FAQ. The result is an exciting analysis of current processors that reflects performance in 2022. The multi-core utilization was primarily taken into account, since even more CPU cores are to be expected by the end of the year. The biggest innovation on the test platform for processor benchmarks is once again the graphics card. Instead of an Asus TUF RX 6900 XT, we now use an RTX 4090.
We haven’t changed anything in the proven test methodology: Before the benchmark runs, each CPU is subjected to a warm-up period, during which the core clock (boost) drops to a specific level that varies from model to model. We configure a TDP limit in the UEFI of the test mainboards: Each CPU is configured to the parameters that the manufacturer provides for the respective model. With an Intel Core i9-10900K, for example, this is 125 watts (TDP), while the boost can apply 250 watts for 56 seconds. By heating up, we force each CPU to go into its sustained load mode. In this way we avoid unrealistically high values and enable a fair comparison. The clock frequency of the installed RAM is also forced to the manufacturer’s specifications, we specify the amount of 32 GiB (DDR4 and DDR5) and the timings (XMP).
We always use the latest version of Windows 11 including all (security) updates, but without TPM 2.0. Each benchmark consists of three runs, each lasting 20 seconds. As before, the lowest possible 16:9 resolution is used for the benchmarks; sometimes we reduce the resolution even further using scaling to minimize the GPU load. We use the current version of CapframeX (CX) as the only tool for recording and evaluating the average performance achieved (average fps) and the frame times (percentile P1 (milliseconds in fps)). CX enables outlier detection, which we apply to the 1% frame times with a maximum of five percent, which increases the precision of the results.