Couple things. While it's implied, there is no actual statement as to what cooling was used in either of the i9 tests. Stock cooling? Explain.
They did at the very beginning of the article, & even indicated that they were air-quoting "stock" for both the 3900X & 9900K more correctly meant "out of the box" (I.e. only the necessary XMP profiles needed to run the RAM at the RAM's stock speed, the stock cooler on the 3900X, & the Corsair H115i cooler used for the 9900K but without any manual changes to clock speed).
How many times was each benchmark ran? Was this an average of results? Was this using the same parts except what was stated? Was it an open test case? Were they performed at the same time?
Although not mentioned in this particular article, they have mentioned in enough articles over the (many) years of their reviews that they do multiple passes through the same benchmark on each game so that they get an
average result. Guess they assumed by now that people would remember that.
Is 1080p considered a benchmark anymore? Why use a 2080ti for 1080p? Can we get cpu and gpu (gpu especially) usage percentage scores (high/low/average)?
Jesus, people act like it's tough to properly report findings. This just comes off as sloppy, I expected better, honestly.
They go into it in much more depth in an article from last November (https://www.techspot.com/article/1637-how-we-test-cpu-gaming-benchmarks/). However, it's a very long article with a lot of different examples, so I'll summarize:
-- There are 3 ways that a game/system combination can be bottlenecked: CPU, GPU, & a combination of the two.
-- The higher the resolution, the higher the chance of a
GPU bottleneck occurring...& even with the mighty RTX 2080TI, there are still games that at 4K resolutions that will be GPU bottlenecked. It's simply that the GPU takes so much time to process & display the data it gets from the CPU that the CPU spends more & more time idling while it waits for the GPU to catch up. That's also why there are a number of games where, at 4K resolutions, pairing
almost any CPU with the "wrong" type of GPU will show zero difference in performance.
-- The lower the resolution, the higher the chance of a
CPU bottleneck occurring. In this case, even a mid-range GPU (let alone a high-performance one) easily blows through the data it's received, so now the GPU is waiting for the CPU to send it some new data.
-- They used to do CPU testing at 720p or lower resolutions, because only high-end GPUs could provide 1080p (or even 1440p) output. Nowadays, 1080p is considered the "normal" or "standard" resolution (Note that in both the recent May & June Steam hardware surveys, 1080p was the most common resolution used by gamers; of the top 26 GPUs from May, the top one was the GTX 1060, a 1080p GPU, & out of those 26 GPUs only 6 of them can be considered 1440p GPUs, let alone 4K, with the other 20 GPUs -- essentially being "1080p or lower", including the integrated Intel options -- providing 52.95% of the GPUs in the survey; the article on the June survey only included the top 19 GPUs, but again only 6 can be considered 1440p/4K cards, with the other 13 still providing 48.93% of the hardware out there), as a good mid-range GPU can still be bottlenecked by a weak CPU.
-- The last thing you want when testing is to use a system that's bottlenecked on both the CPU & the GPU. So, in order to maximize your ability to measure differences in the performance of various GPUs, you
minimize any effects due to CPU performance by using the most powerful CPU available (or at least close enough for the majority of users), & test at the available resolutions to show how their performance scales; conversely, to measure differences in performance between different CPUs, you use the most powerful gaming GPU available (in this case, the RTX 2080TI), using a fixed resolution, & then optionally testing to see how overclocking affects the performance.
-- Expecting them to test all conceivable combinations of CPU, GPU, & resolution across all of the different motherboards, RAM speeds, & RAM brands is a complete pipe-dream. But hey, if you have unlimited cash & time on your hands to do it yourself, more power to you.
TL-DR: It's been explained for years & years, not just here but on all of the
reputable benchmarking sites, why CPU testing is done this way. It's done that way because that's how things work. Otherwise, any "results" from the testing are essentially meaningless.