This topic is actually very simple. Rockstar Games based Grand Theft Auto's original 3D trilogy upon a modified version of 'RenderWare', a game engine developed by Criterion. According to
this post by RenderWare developer 'Simagery', the game only uses the 3D graphics engine of RenderWare, but uses Rockstar's own systems for everything else, including physics. CPUs use hardware-based oscillators to calculate time, and that information is fed back to the operating system. The original trilogy used PS2 as its lead platform, targeting a 33.3ms frame-time (30fps). Under these circumstances, Rockstar knew that the maximum frame rate was 30fps, and they knew that the game
shouldn't drop too far below this value. As a result, most of the game logic and physics was set to operate at a fixed time step of 33.3ms. However, major aspects of the game logic and the physics were calculated using a variable time step, meaning that the game could adapt to variances in frame rate. This was mostly fine for RenderWare's graphics engine, but many aspects of the handling and physics were calculated more quickly as frame rates increased, resulting in issues like different friction, braking forces, rotation speeds, and more.
The difference with GTA San Andreas is that the game had an additional 14ms
delay between frames on PC, which limited the frame rate to 25fps. This additional frame delay persisted when Vsync was disabled, and prevented the game from rendering unlimited framerates with Vsync disabled (FL-OFF). The problem is that this frame delay was core to the physics of San Andreas, because the game also needed more
time to complete physics calculations. When you use modifications like SilentPatch - which remove this frame delay to restore the Vsync lock of 30fps, and allow frame rates up to an artificial cap of 1000fps - physics calculations happen more quickly at any given frame rate, and this results in the types of physics glitches you see when playing with these mods enabled, or playing MTA/SAMP, which both remove this delay themselves (albeit with different FPS limits).
Outside of frame-times, the only difference with San Andreas is that the replay system captures data at a fixed rate of 40ms (25fps), whereas Vice City captures data every time a new frame is rendered. San Andreas drops frames so that it can always capture the same amount of data per second, and make sure replays playback identically regardless of frame rate. Vice City suffered from variable frame rates within replays, so when you played them back at 30fps, they would play back more slowly/quickly depending on how much data had to be played back per second. Frame drops would result in faster playback, since it would take less frames to move the same amount of distance, and high frame rates would result in slower playback, because it would take more frames to move the same amount of distance. This also resulted in drastically shorter replay lengths, because more data had to be fit into a limited amount of 'blocks' stored in RAM. As Daffy mentioned, people
could have instigated their own 60fps cap to allow for high FPS videos with consistent replay playback, but frankly, nobody did this because there was no demand for it, so the point is essentially moot. Of course, that doesn't stop anyone doing it now; it might be cool to see a video made with a 60fps cap, and past precedent is not a good reason to shy away. I don't think it's a good idea to change the
standards now though, because they're not broken.
This highlights the central issues people have with FL-OFF in Vice City. Frame rates are uncapped, so physics calculations are far faster than San Andreas, and Vice City's replay system doesn't drop frames at higher frame rates, so replays invariably look different on different systems, and the length of replays isn't good enough. However, what people consistently fail to realise is that almost exactly the same things happen when you use FL-OFF in San Andreas, barring Vice City's replay system issues. Why don't people understand this? The differences simply aren't as obvious, because San Andreas has an additional 14ms of delay in physics calculations, and frame rates are locked at ~71fps without modifications (the frame-times aren't 100% accurate, so you may find small deviations).
- Most people seem to run mods that cause the game to max out around ~142fps instead (7ms), but I'm not sure which modifications specifically cause this; perhaps it's to do with No-CD patches, or Windows, or AMD/Nvidia's drivers? Would require more testing than I have time for. - SilentPatch is perhaps the greatest demonstration we have of this, for it shows how San Andreas' handling and physics operate under similar conditions to Vice City's runtime. That is why people vehemently stick to 30fps in Vice City, and why using SilentPatch is considered modding in San Andreas. Now as for why people use FL-OFF in San Andreas... different people come to different conclusions, and the negative effects of San Andreas' unlocked frame rates were not as apparent when the game was released, especially since computer hardware was far less powerful in 2005, so many people didn't even realise there were any differences until they discovered the RAD method, by which time they were already used to FL-OFF.
Should anything be obvious, it is that FL-OFF in Vice City is not "modding", it is simply looked down upon, because those who use it have a significant advantage over the stunting community as a whole, who have all stuck to FL-ON. San Andreas stunters may well have done the same, had they understood the consequences of FL-OFF from the start. If you want to stunt in the same conditions as Vice City stunters, you should stick to FL-ON, and for San Andreas, you should usually stick to FL-OFF, outside of circumstances where FL-ON is beneficial. As for what remains
unknown, we don't have access to the source code, where we could definitively see how each individual function is calculated. Without that, all we can do is research the effects of FL-OFF within the game itself. As a community, we have a pretty good understanding of what that means, but in order to document the
exact differences, that would require tangible efforts on the part of those whom it concerns. I don't see that happening right now, all I see is the playground attitude that pervaded the stunting community long ago.
Disclaimer: I'm not a programmer, I just did some research about this back in the day. I might be wrong about some things regarding fixed/variable time steps, for I'm working with other people's documentation, and I have never done any programming, so I can't practically test my findings. If you want more detail about fixed/variable time steps, and their relation to game logic or physics, ask somebody else.
