Perhaps the most popular benchmark in reviews and CPU performance measurements is Cinebench, which simulates 3D rendering in Maxon’s professional Cinema 4D software and can be used as a metrics for comparing CPUs. Since version R23, it also benchmarks Apple Arm processors. Maxon has now released a new version, Cinebench 2026, which works better with the latest processors and GPUs and also includes an interesting new mode.
Cinebench 2026 follows the version 2024 released two and a quarter years ago and is based on Maxon’s Redshift rendering engine, which has once again undergone further development since the previous version. The benchmark is available for Windows 10/11 and Apple’s macOS as before, with both operating systems supported in x86 (or better said, x86-64) as well as Arm versions.
The new version uses different code and also a different compiler version for building the binaries, so once again the scores cannot be compared with those from older versions (and the scale is also completely different to avoid confusion). The test also renders a differently configured scene. Maxon states that it should be roughly six times more demanding in the multi-threaded test, which makes it ready for increasing core counts in future CPUs (for example, Intel is preparing desktop models with 52 cores, and AMD will also be adding more cores to future AM5 processors). So be prepared for scene completion to take longer on today’s processors than in the older version.
As with previous versions, the program by default renders the scene multiple times until the runtime exceeds ten minutes, and only then calculates the score. This is done to capture the behavior of processors that may deliver maximum performance only for a limited time and then slow down—a common occurrence in notebooks. This behavior can be manually disabled, however, allowing the test to run just a single standard pass.
Measuring the benefit of SMT
What is completely new in Cinebench 2026 is the addition of another results category. From the main screen, it is now possible to directly launch a test calculating a score using a single CPU core, which earlier versions only provided after manually selecting advanced testing options. It is no longer hidden.
A bigger novelty, however, is that Cinebench 2026 directly attempts to measure the benefit provided by SMT technology (called HT—Hyper-Threading—by Intel). Until now, this technology was used in the multi-threaded test, but was logically ignored in the single-threaded test. However, this obscured the benefit SMT can bring—if you understand a “single-threaded” test as something intended to indicate the performance of a single core, then you should really be interested in performance with SMT enabled (that is, using two threads, or theoretically more in the future if processors begin providing more threads per core). Cinebench 2026 therefore now measures and scores both modes.
You can thus compare the difference between a core rendering with only one thread and with both threads engaged, which is labelled as “single-core performance.” It turns out that when measured this way, the performance of Zen 5 cores, for example, is significantly higher than would be implied by the relatively low single-threaded scores this architecture receives in Cinebench tests for some reason.
It cannot be simply said that one of these scores is correct and the other wrong. At least for the purpose of practical evaluation of performance for “productive workloads,” it makes sense to compare single-core performance (which means one thread for CPUs without SMT and, for example, two threads for cores with SMT). The difference is far from small—on the PC we used set up with AMD Ryzen 9 7950X with Zen 4 architecture, SMT added almost 31% performance when running on a single core.
If you are not interested in practical performance and instead want to use the score as an indicator of how many instructions and memory operations an architecture can execute in a single thread (which relates to so-called IPC), then you may continue to prefer the pure single-thread “1T” score, even though it has more of an academic or theoretical significance—in practice, users of CPUs with SMT will certainly enable this technology when rendering in Cinema 4D.
Still won’t run on old CPUs
The test has similar requirements to version 2024. On x86 processors, this means AVX2 support is required, so the oldest CPUs that will work are AMD processors with the Excavator architecture and Intel Haswell (4th-generation Core). Under macOS, where Intel processors are still supported, this requirement does not apply—SSE4.2 is sufficient.
Windows 11 on Arm processors is also supported. These must support the ARMv8.1 instruction set, and the program officially supports Qualcomm Snapdragon chips.
The computer must have 16 GB of RAM; the exception is on macOS where the test can be run on 8 GB or 12 GB configurations, but performance will be heavily degraded due to swapping data to disk. On Windows this is not allowed, but this does not mean that macOS uses RAM more efficiently. According to the release notes, the reason is that Windows reserves a larger portion of RAM as non-pageable, so Cinebench does not fit into the remaining memory, whereas macOS is able to swap a larger portion of its memory to disk. Scores with less than 16 GB of memory, however, will—as mentioned—be reduced and will not show full CPU performance.
GPU can be utilized as well
Cinebench 2026 once again supports GPU testing. This test requires 8 GB of graphics memory on a discrete GPU; for Apple M1 through M5 processors with integrated GPUs, 16 GB total RAM is required—unlike the CPU test, 8 GB or 12 GB is not sufficient. Supported graphics cards include AMD Radeon GPUs with Vega or RDNA architectures and newer, and Nvidia GeForce GPUs with CUDA 5.0 support, which should include Maxwell (GeForce GTX 900), Pascal (GTX 1000), and newer GPUs. On macOS, only integrated GPUs in Apple processors (generations M1 through M5) can be tested; previous support for discrete Radeon GPUs has been removed for this operating system.
Available for free
You can now test your processor yourself—the benchmark was released yesterday and is once again freely available at no cost. You can find it on Maxon’s website here.
If you want to show off your tuned systems and their results, feel free to do so in the discussion. The system shown in the screenshots above is running at stocksettings (except for EXPO profilee being enabled for the DDR5-6000 modules), and the CPU is already three years old at this point, so you will likely outperform it fairly easily if you have a processor with a slightly higher core count.
English translation and edit by Jozef Dudáš
⠀
This article says the “Cinebench 2026 is here.” Where is “here”? Where do I find these 2026 scores?
We’d be happy if you shared your results in the comments on HWC. When the article says “Cinebench 2026 is here,” it simply refers to the application being available. 🙂