It actually doesn't read from RAM at all, or read the controllers. It's all ROM. There is a bit of "uninitialized RAM shenanigans" for the 9 cpu cycle run to work, so I also put together a 30 cpu cycle run that works from the moment the console is powered on. for the 30 cycle run, Assume RAM is in the console's resting state, and the contents of the MMC3 chip are nondeterministic (That needs to be initialized before the credits can roll). And then I made a similar run for SMB1. Very excited to get that video out, but I've still got a lot of editing to do.

The first step I usually take is hopping in the game and using emulator tools to search for what addresses change when I'm modifying the address I'm looking for. For instance, you could search for what bytes change between picking up a mushroom or fireflower. The actual method I use is a bit more advanced, but with Mario 3 specifically, I just look at the assembly code. There's a really well documented disassembly, and I can usually just CTRL + F for whatever variable I'm looking for and find the address pretty quickly.

@@100thCoin Do you have a link for the disassembly?

Yeah, you could definitely use arbitrary code to make a playable tetris.

I am certainly a nerd.

without the domestic abuse :(

Sorry about the background! I'm currently working on the next video, and I'm making sure the background is less busy. And yeah, the NES is capable of reading inputs this fast! The real limitation is providing inputs at that speed. Using modified controllers, this TAS has been run on real hardware. Source 1: kzbin.info/www/bejne/p5LXppKmrdSXjK8 Source 2: kzbin.info/www/bejne/gKqrc2R4fMx3bqs The "Bad Apple TAS" that I made requires inputs around 4 times faster, which has also been done on real hardware: kzbin.info/www/bejne/hXu7qaB6lLaeaNE

Using high energy particles to flip bits would require some sort of external "high energy particle source" that likely wouldn't be licensed by Nintendo. With the 5 microsecond run, it could (in theory) be performed using official hardware, which I think makes it a bit more interesting.

@@100thCoin Thank your for the reply! Needless to say, a speedrun needs to be done on unmodified hardware. However, the tools needed to playback the input sequence do not need to be liscensed. One could imagine to play back inputs by means of having small, computer-controlled motors pressing buttons on the offificial controller to play back a sequence of inputs. Slightly more advanced, you get a thrid-party connector and that way send electrial pulses directy to the NES. The game still runs on official hardware. How, the NES is a physical device. While it is designed to accepts button presses as inputs, it is sensitive to physical phenomena. You may have heard how a a cosmic ray, i.e. ionizing radiation in form of a proton has flipped a bit in memory during a stream of SM64 which caused an upward. Silicon-based computer memory is generally sensitive to ionizing radiation. Thus, the unmodified NES too will be sensitive that too. So, we come to my idea: How about we just flip all the bits of memory we need to have cleared the game by means of ionizing radiation instead of waiting ages for the CPU/memory to tick? That way, we can clear as fast as we can produce a way to manipulate memory. If done by optical means or by high energy particles, this can be faster than electricity. The thing is, humanity can manipulate ionizing radiation with great detail. There is a good chance that the computer you are using right now has been made by a very precisely formed pattern of ionizing extreme ultra violet light (Keywords: EUV photolithography, 4 nm process node) . Hence, the truly fastest way to possibly clear SMB3 (or any computer game, really). Directly manipulate memory with EUV light pulses.

@@TheTablet314 If you build a device to do that, I'd love to watch a video explaining how you put it all together! For my own personal goals though, I think I'm satisfied with the method I'm using. Though this talk about external tools makes me wonder about how I could emulate the one hotplate% speedrun. Imagine making an emulator that supports a simulated hotplate under the console, ha!

If you had an insanely fast computer capable of converting the screen into mario tiles, then compressing those tiles into packets, then running the frame to see if the $2002 race condition needs prevented, then converting it into button presses and sending it to the replay device all within 1 frame, then yeah- probably.

It's actually 7-bit PCM. It's just being played at 25 Kilohertz.

@@100thCoin 7-BIT?

@@Rory69_ Yeah, the "Delta Modulation Channel" of the audio chip can either play 1-bit DPCM samples, or 7-bit PCM samples. www.nesdev.org/wiki/APU_DMC

Delving into assembly programming is very fun and interesting. Maybe it was just over your head :)

TASers should stop doing what they love because you aren't interesed in that? You realize that what's interesting or not is a subjective thing?

With an unmodified controller, you're probably going to damage it by having a robot actually push these buttons at the speed required to make this run work. I don't think it would break the plastic shell or the board inside it, but I wouldn't be surprised if the buttons stop working.

Yeah, I probably should have.

For 6502 Assembly, the $ prefix is used instead of 0x.