Intel Core i9-10900K processor review - Performance - CineBench 15 (and IPC)

Challenges of testing a processor

Each year it is getting increasingly complicated to accurately test processor performance, the motherboard manufacturers are in control of a lot of settings. Intel, on the other hand, does not release reference review motherboards anymore, thus motherboards manufacturers will do tons of tweaks 'at default' in the BIOS just to stay ahead of the competition and tho spread their lineup relative for performance from mainstream towards premium. Settings can include: All cores frequency locks, clock tweaks, and enhancements in power states and duration. For a motherboard review, this is all fine, however, for a processor review, we want to test as close as possible to reference performance. To bypass this challenge, we opted the ASUS ROG Maximus XII Extreme for this reference tests, as the BIOS offers a setting that enables reference Intel Turbo ratio settings, see below:

You expect you will see lots of reviews with higher performance numbers as many media will load up BIOS defaults and start testing without checking. We thus test at reference performance (or as close as possible) with settings as referenced by Intel. Before anyone asks, yes ASUS in the premium models also offers an all-core enhancement, this is set manually to off as well. For motherboard reviews, we, of course, revert to motherboard defaults as it is well within their right to segment and optimize their lineup performance-wise.

Processor performance: CineBench 15

CB15 supports systems with up to 256 threads. The performance of processors and graphics cards is, as usual, determined on the basis of 3D scenes. A selection of test results allows a rough classification of the benefit of your own system. For the CPU test is a scene with around 280,000 polygons used, while the GPU test based on OpenGL comes with about a million polygons, high-resolution textures, and various effects. The results will be issued in final points (CPU) and fps (GPU). According to the developers, the software has been "extensively developed to exploit the performance of new hardware as possible." The results are unsurprisingly not comparable with those from earlier versions. You'll notice we still need to add a number of processors, all in due time. You'll notice the single-core perf paints a completely different picture here.

Instructions per cycle (IPC)

This IPC test will build up and get updated over time. We lock all processor cores at 3500 MHz. That way you can see the architecture performance of the processor clocked at exactly the same frequency. This is a single thread measurement. For many people, this is the holy grail of CPU measurements in terms of how fast an architecture per core really is. I, however, tend to say there's more to it than that and that would be higher frequency allowances, caches, and memory latency defining that per-core performance.