They only go so far because the clock speeds in RDNA 2 GPUs for instance in the 6800 and 6800 XT are already well over in the 2250+ MHz region, they hit diminishing returns when pushed past their factory clock speeds.
In the console space, XSX operates at just over 1800 MHz, which is not "marginally" less, that's very, very less by RDNA 2's standards and almost 500 MHz slower compared to some of the desktop parts. Anyway, PS5 GPU operates at 2230 MHz. This a massive 400+ MHz (22%) difference here, we aren't seeing this kind of difference in clock speed between different RDNA 2 tier GPUs on PC (only around 80MHz or so between 6800 and 6800 XT at stock for e.g.).
The best way I can explain: the situation between PS5/XSX GPU is akin to someone manually downclocking a higher CU RDNA 2 GPU (e.g. 6800 XT = XSX) and keeping the lower CU RDNA 2 GPU (6800 non-XT = PS5) at stock clock speeds. Performance characteristics of these GPUs when you do something like this will most likely result in some games performing better on lower CU GPU, some will be on par/close/similar (as it has been with PS5/XSX in many games now) and some that really favor CUs will perform better on higher CU part. That's the kind of situation PS5 and XSX are in.
Saying that no game favors higher clock speed would also be false because we've seen in many games PS5 outperforming XSX or have similar perf. RE8 is the most recent e.g. of this. As I've said in my prev posts - perf will flip-flop depending on scenes.
If higher CUs were the only way to go and we can't expect a higher clock speed on fewer CUs to have similar perf -- then this shouldn't be happening (same graphics settings/4K CB/no dynamic res):
When benchmarked the avg. fps between PS5 and XSX difference is in 0%-2% range, do you know?
Notice the similarity in the graph line. This isn't some fluke. This just points to how smartly the PS5 is designed it's achieving similar perf as the 12 TF console with less silicon real estate.
This isn't really a good comparison because GPU manufacturers carefully tweak and tune their GPU lineup so that lower-tier GPU doesn't outperform higher-tier GPU even when OC'd. You can't take this logic and apply it to MS and Sony's console GPUs.
I know what you mean and I've done extensive research on these architectures (Turing/Ampere/RDNA 2...), their lineup and down to each subsequent GPU's configs: SM/CU/ROPs/TMUs, primitive units, etc. and have come to know what kind of rasterization, fillrate throughput each of these and console GPUs have at their clock speeds.
Let's take your 3090 vs 3080 example.
According to Gamers Nexus:
RTX 3090 plots: 1890-1905 MHz
RTX 3080 plots: 1920-1935 MHz
3080 indeed has a higher clock speed. But... Did you care to look deeper? I'll take 3090's min clk speed and 3080's max clk speed for the comparison below:
3080 =
1935 MHz +2%
Pixel fillrate:
186 Gpix/sec -12%
FP32 TFLOPS:
34 TFLOPS -15%
Texture fillrate:
526 Gtex/sec -15%
3090 =
1890 MHz -2%
Pixel fillrate:
212 Gpix/sec +14%
FP32 TFLOPS:
40 TFLOPS +18%
Texture fillrate:
620 Gtex/sec +18%
You can see every aspect of the 3090 GPU is faster than 3080 despite 3080 having a slightly higher clk speed. This is what I was talking about when I said GPU manufacturers carefully tweak and tune their lower and higher tier GPUs so they perform as they intended. Those 14-18% gains across the board clearly translate to better gaming perf for the 3090 because that is intentional.
Now look at PS5 vs XSX GPU:
PS5 = 2230 MHz +22%
Pixel fillrate:
143 Gpix/sec +22%
FP32 TFLOPS:
10.3 TF -15%
Texture fillrate:
321 Gtex/sec -15%
Rasterization rate:
9 Gtri/sec +22%
XSX = 1825 MHz -18%
Pixel fillrate:
116.8 Gpix/sec -18%
FP32 TFLOPS:
12.15 TF +18%
Texture fillrate:
380 Gtex/sec +18%
Rasterization rate:
7.3 Gtri/sec -18%
Here not all aspect of the XSX's GPU is faster like the 3090 was, as you can see. That 22% increase in clock speed over XSX is so substantial that some parts of the PS5's GPU receive that much higher throughput over XSX's GPU leading to similar perf (proof above) or in some cases better perf for the PS5. You're just looking at CUs and TF and claiming that it's all that matters when it's not. If it did, XSX would come out on top in each and every comparison consistently, and not have similar perf. Let's appreciate both the designs for a second and enjoy the games.