Emulator

I forgot to mention I did most of the testing using Vulkan and nVidia's 445.87 drivers.

I've been doing some more testing with Pokemon Colosseum. All backends initially behave similarly. However, the stuttering is far less severe with both D3D11 and D3D12, ends up going away after around 5 minutes or less and isn't there on subsequent boots. From what I remember, neither had issues with the title screen and only started stuttering like crazy after loading into a map. On OpenGL and Vulkan the stuttering is far worse, starts at the title screen and is still there if you close and boot up the game after it goes away. I thought it might be related to shader compilation but I'm using Async Ubershaders and I have Compile shaders before starting on so I don't think that makes a lot of sense.

Does toggling "Compile Shaders Before Starting" change anything?

It doesn't seem to have any effect on Vulkan or OpenGL. By the way, OpenGL doesn't start stuttering at the title screen now, it starts when you load into a save. I'm starting to think either my shader cache is corrupt or there's something messed up with my windows install, since I tried booting from the other computers' hard drive and dolphin worked fine with (as far as I could tell) the same settings, same drivers and I think same updates. The two biggest changes with that would be the windows install itself (mine might be corrupted but the infamous SFC utility can't find anything) and the shader cache from the other Dolphin. I don't have time to do any more testing now but I'll save a copy and see if deleting the shader cache does anything.

To test if it's the shader cache being corrupt, nuking everything inside the Cache folder should do it, right?

So, I've been trying more stuff and I've found out this only happens in 60Hz mode. However, F-Zero GX doesn't have this issue, regardless of what screen mode you choose. I don't think Colosseum and Metroid Prime share the same engine but they might share the audio microcode, in which case, does F-Zero GX use the same audio microcode? I don't quite see how this would affect framerate without affecting the emulation speed itself but I'm just trying to narrow down the specifics of this.

Skip Duplicate Frames + Dualcore is buggy. I don't believe it's possible for the audio to be the culprit, but for what it is worth, those games all use the AX microcode.

It's only really buggy with those two specific games (though I should test other AX microcode games) and only in 60hz mode. What ticks me off is that it doesn't seem to happen on Intel CPUs, but then again it might be down to IF shenanigans with Zen. It would be interesting to test whether a 3300X would behave the same but that's besides the point. Anyway, I've been doing some more testing, I'll write here some of what I've found just in case.

Both Sync GPU and Deterministic dual core fix this without having to give up all of the performance gained with the dual core hack. For what it's worth, it might be useful to have either of these enabled in the game-specific settings by default or to add a warning about this on the wiki. Also, with the chipset drivers the issues are far more severe and inconsistent so go figure. This might also explain why booting from the other computer's drive seemed to fix the issue, since it didn't have any AMD chipset drivers on it and without them only the world map seems to have this issue (I was testing on Colosseum for the most part). I've also noticed that the pipeline cache file is about 3/4 the size when generated using the drivers. It seems newer versions of Windows 10 come prepackaged with an AMD chipset driver so I had to install 1703 to test without drivers.

Oh, I knew I was forgetting about something. With the drivers, performance is very inconsistent. Sometimes the game ends up running at about 20 FPS. When that happens, the player character seems to behave in a way that reminds me of the "Shaky props" issue in Mario Sunshine. Combined with the fact this only happens on Zen, with the dual core hack on, running at 60Hz and with Skip uplocate frames on it might be down to IF being far laggier than the ring bus coupled with something those two specific games are doing.