Trying to find my performance bottleneck on MGS3

[cyanea]

Hello, I'm getting some slowdowns in MGS3's opening cutscene and would like to find out what exactly is the limiting factor here.
Slowdown always happens at the same 3 scenes for about 2 seconds and outside of those 3 scenes the opening cutscenes do run at full speed.

Increasing EE Cyclerate seems to do nothing and going by the OSD details, at least to me it doesn't look like a EE limited scenario anyway.

Lowering Internal Resolution to 2x resolves the issue completely, while increasing it to 4x makes the slowdown worse (down to ~45 fps, instead of ~54fps on 3x Res). Looking at usage in task manager, my CPU and GPU seem to experience only light load. Increasing core clock on the GPU seems to do nothing, while giving +1000Mhz to the memory clock seems to improve the issue slightly. Though that could just be confirmation bias on my end. 

So if I had to make a guess, I'd say that I'm limited by memory speed or bandwidth on the GPU. But I'm all ears for other ideas 

CPU Ryzen 5800X
GPU GTX 1060 6GB

Renderer is set to Vulkan, settings which differ from stock are:
Anti-Blur - Off
VSync - On

Internal Resolution - 3x Native
Anisotropic Filtering - 16x
Blending Accuracy - High

60 fps patch - enabled

I'll attach some screenshots with OSD enabled.

[refraction]

readbacks are probably our issue, you can try disabling them but it may cause some graphical errors.

IF you go in to the per game settings while the game is running, then Graphics->Advanced, there should be an option to "Hardware Download Mode", set that to anything but "Accurate"

If that doesn't fix it, try lowering your blending accuracy, or resolution.

[cyanea]

Alright, I looked at GPU usage during these slowdowns in MSI Afterburner and the GPU load does indeed get to 60%.

MGS3 needs blending at high, can't really lower it. Tried out all Download Modes but setting to anything other than Accurate results in graphical errors. Though setting it to "Disable Readbacks" does indeed get rid of the slowdowns.
So I guess the solution for playing at 3x/4x native would be to buy a better GPU? A RTX 4070 for example has about 2x my current effective memory clock and ~2.5x on the memory bandwidth (and is obviously a much more powerful GPU in general). Or would that not improve anything in this scenario?