If you refer to Jokers review, should maybe turn on the sound. He explains why the workloud is 99% in his 1080p tests. VSYNC off means gpu load will always show 99%. He discusses it further in his follow up videos.
It is incredible frustrating to see people trash reviews without an idea whats going on.
I did watch the video with the sound on.
His 'explanation' that disabling V-Sync allows the GPU to hit 99% load shows a lack of understanding of the issue.
When you set up a CPU-limited test, the GPU
cannot reach 99% usage.
If his tests are set up in a way that allows the GPU to hit 99% load, then performance in the game is being limited by the GPU - it cannot work any harder than that.
Therefore if the test is always at 99% GPU load, it's measuring GPU performance and not CPU performance.
If it's bouncing around between 90-99% for example, then the test tells us nothing useful at all, because that test is GPU-limited in some scenes (99% load), and CPU-limited in others. (<99% load)
A CPU test must be set up so that the GPU
cannot ever hit 99% load.
I would try to keep it
at least below 90% at all times - ideally much lower than that.
And the easiest way to achieve that is to use the fastest GPU that you have, and run it at a low resolution.
It doesn't matter which GPU you use, as long as you use the same GPU and settings with all CPUs that are being compared, and so long as it never reaches 99% GPU usage when testing with your fastest CPU.
When the GPU is unable to hit 99% usage, it generally means that the CPU is limiting the framerate, instead of the GPU.
That is how things need to be set up when you are trying to benchmark the CPU.
-----
Here's an example of CPU-limited results in
Deus Ex: Mankind Divided on my PC. In all tests, CPU usage is at 100%.
Now that's actually quite rare - most games won't load up the CPU to 100% usage on all cores, but I chose it for this example because it makes it clear that the CPU is the limiting factor.
In this test I am using low enough settings that even a GTX 960 is not coming close to 99% GPU load.
When I swap the GTX 960 out for a GTX 1070 - though it could be
any faster GPU - the only thing which changes is that the GPU load drops from 60% to 33%.
Framerate remains the same despite the faster GPU, since the CPU is the limiting factor.
The faster GPU does allow me to turn the settings all the way up and have essentially no impact on framerate though - it drops from 43.7 to 43.3 FPS.
⠀
Now what would happen if I used a faster CPU for these tests?
I don't have a faster CPU here to show actual results, but I can tell you what the results would be like.
In the first two tests, a faster CPU would increase GPU usage, which would allow for higher framerates.
If the CPU is fast enough, GPU usage would increase to 99% usage on both cards.
If we assume that everything scales linearly - it likely does not, but that assumption keeps the example simple:
Since the 960 is at 60% load and 43.7 FPS, we could say that a fast enough CPU would allow the game to run at 72.1 FPS when it hits 99% GPU usage.
With the 1070 at 33% load in the 720p test, a fast enough CPU might allow it to run at 131.1 FPS using the same settings.
The 1620p test is already GPU-limited though, with the settings tweaked so that it's essentially running at 100% CPU and GPU usage at the same time.
The only thing that a faster CPU would do for this test is drop the CPU usage.
Performance would remain at 43.3 FPS since the GPU is already maxed out.
Now if I was trying to benchmark a group of CPUs, I could not do this at 1620p on a GTX 1070.
All the CPUs in the test would show a 43.3 FPS result, unless they were slower than my 2500K and caused the framerate to dip even lower.
CPU A might be able to run the game at 60 FPS and CPU B might be able to run it at 90 FPS, but you'd never know because the test is GPU-bound and both will show a result of 43.3 FPS.
A CPU comparison would have to be done using the 720p settings - or possibly even lower - to prevent the GPU from ever hitting 99% and affecting the results.
If the settings cannot be reduced any further and the GPU load is still at 99%, then we need to swap out that 1070 for a faster GPU.
They addressed it on Reddit. They simply wanted reviewers to test at all resolutions, not just 1080p, to give a full picture of the performance across resolutions.
The only thing increasing resolution does is increase the likelihood that you will be GPU-bound again, which compresses the results and unfairly narrows the gap between the faster and slower CPUs. (in favor of the slower CPU)
