Jump to content
IGNORED

Todd's 5.51 Dragster score


homerwannabee

Recommended Posts

Omnigamer, thank you for all your excellent work! And welcome to AA, hope you stick around.

 

After spending all weekend reading the fascinating dispute thread on TG forums, I am humbled and amazed at your patience in dealing with Jace. It's quite obvious that he has an agenda and brand to protect, and it is painful to see played out. It's like trying to debate a flat-earther. Let me assure you as a simple Atari lover watching from the sidelines, don't give up the fight!

 

From all I've seen, there's no way Todd made 5.51 legitimately (either accidentally or intentionally). Like others mentioned, I think the wisest solution (since removal seems almost out of the question) would be to asterisk unvalidated and likely impossible old records like Todd's Dragster record.

 

The only thing I can add, which I saw someone else mention, is on an actual console + cart, it is possible to "fry" games to make them behave differently by turning them off and back on really quickly. As a child, I found this out accidentally with 2600 Space Invaders and there was a certain screen I would wait for that would allow me to shoot multiple shots (as opposed to waiting for each shot to connect or go offscreen to shoot again), which made the game much easier and more enjoyable. It usually takes 10-30 power cycles, and it is most definitely repeatable. Perhaps Todd found some trick like this. It would explain the secrecy and the lack of simply streaming his attempts on Twitch.

Link to comment
Share on other sites

I don't know the exact effects from repeated power cycling, but like I stated earlier it seems likely to me that it's leaving the program counter or other registers in a dirty state. This could do a lot of things to program flow and execution, but if you're at that point then you're already not playing the same game as far as I'm concerned. I lump that in with my blanket case of "hardware failure."

Link to comment
Share on other sites

I agree completely. I just mentioned it because (when frying Space Invaders) after I got the screen I knew allowed for the multi-shot, I could press Reset to start and was able to play as many games as I wanted with multi-shot even after pressing Reset. It would stay "fried" until I turned it off again. The game otherwise looked exactly the same. Honestly I was trying to play devil's advocate with really the only possible way I could see Todd actually achieving that time on an actual console/cart. It could have been "fried" even without his malicious intent or knowledge if he was turning it off and on quickly while making attempts. I'd certainly buy that over the "human element" idea anyways :)

Link to comment
Share on other sites

Why don't you use your talents to show how it was done as it clearly was, instead of trying to prove that it is impossible on the computer? Activision's computer simulation already said that 5.51 was impossible, so your computer evidence is nothing new. Todd proved them wrong long ago, multiple times. Since you are quite talented on the computer, maybe you can show on the computer exactly how Todd did get a 5.51.

The current models prove that 5.51 is impossible based on them. Since they represent the way everybody (except Todd?) is playing the game, it became obvious that this could be a reason, why a score of 5.51 looks impossible. So either there must be something missing in our models and play approaches (which might explain why only Todd got the current record) or the record time is no true.

 

I am pretty sure that Omingamer, just like me, has been and still is looking for flaws in his model.

Edited by Thomas Jentzsch
  • Like 2
Link to comment
Share on other sites

Your first frame is actually the last frame before the time starts ticking. The speed doesn't increase in the 1st frame, because of the initial shift into 1st gear. And every frame, the current speed gets accumulated to distance before the speed is increased.

 

That's why you see the big delay in distance growth.

Edited by Thomas Jentzsch
  • Like 3
Link to comment
Share on other sites

  • 3 weeks later...

It was mentioned here that a hardware run would be different than the emulated perfect run, would it be possible to burn the hacked version of Thomas Jentzsch into a Atari cartridge and run the "perfect button pressed code" on actual hardware. Obvious all button inputs to start the game have to be patched to make it work when the cartridge load on an actual Atari 2600.

 

All the emulations start with 0 and frame 0, and try to hit perfect buttons at each frame. Instead of trying getting the perfect score why not work it out backwards?

This would validate if a score is possible.

 

Start with the said perfect time of 5.51 backtrack frame by frame, on each frame deduct the maximum distance reached and the maximum best time reached before

"the before frame" is finished executing. Do this one by one backwards until all frames are calculated and maximum distance are reached, then see if all deductions

stop short of or go over the zero. That means a score will either fall short or be further optimized. Anyone?

Link to comment
Share on other sites

Sorry, I don't get you here. Can you try to explain again? Moving backwards in time makes everything way more complicated, because one value can be the result of multiple input value combinations from the previous frame.

 

And, if the models are right and there doesn't exist a different ROM, then it has already been proven that 5.51 is not possible. Even a much, much simplified model which ignores all complicated slow downs, doesn't achieve the time.

Link to comment
Share on other sites

What I thought about is start with the desired end results and work out like in a neural network, what is the previous needed input value to get to this value.

Each current frame is the already determined target and the previous frame the possible inputs to get the targeted values.

 

Maybe if you do have a neural network setup, first forget all the permutations just focus in this one layer (the last layer first) and the previous one as what possible inputs

have to be given. The second last frame is the new target and the third last frame inputs the new input layer.

 

So in essence one then do have a backtracked neural network of 330 layers, each layer do have 8 inputs (?) - or is it just 4?.

 

I think doing it this way the possible permutations to get a desired starting target will drop because you do know the target.

 

If frame 1 is reached and the target MINUS desired inputs cannot give the best possible inputs at frame 1 with a starting time of 0 seconds then

the model is proven that a desired time of 5.51 or 5.54 or 5.57 whatever number you start with is or is not possible.

Link to comment
Share on other sites

Ok about the hardware run, some people argue emulation and hardware is so different that the TAS run is different on real hardware.

The only prove doing it run the hacked auto run version on a real Atari 2600 proof to those oaks the outcome is the same in this case.

To do this also mean to write extra code around the real ROM like a "harness/wrapper" the inputs to emulate real input on the real hardware

running the origina code.

Link to comment
Share on other sites

As far as executing known tool-assisted inputs on console, at least two individuals have done this so far. Both noticed no differences between the end state and the expected/emulated state, and ran the experiment several times.

 

As for building a neural net, I think the current models do more than enough to show the impossibility of faster times. Even for brute-forcing optimal patterns, there are more concise ways to cover the search space than to try to back up potential state from an assumed ending value. Formal methods would probably handle that aspect better as well.

Link to comment
Share on other sites

Hi

 

Anyone that do understand the lineup of CPU cycles / score shown vs screen display.?

 

I have a by thought that the only way time display differently than actually achieved that if the actual CPU instruction time versus screen display time do not fall in the same

time interval and that possibly borderline time on CPU time execution will display the previous screen or in other words the previous screen's time will be the time recorded.

 

As one can see below from Omnigamer model, if one add another column to the left to show which CPU instruction slice line up with time achieved and screen display.

It is clear that every 0.0.167 seconds a CPU instruction is executed but only every second one of them every 0.0.334 are used to display time.

 

What if 5.54 is possible and the game's CPU instruction at 330 is executed on the borderline this mean the full instruction set for the screen to display the "4" for 5.54 is not executed.

Will that then imply a 5.51 score?

 

Assumption is made here that CPU instruction nr 1, starts counting at score 0.

 

 

TimeIncr 0.0334 Frames 330.00 Distance 97.94 Dist/Frm 0.30 Hundred of Sec 100.00 CPU cycles / sec 60.00 CPU Instructions / 60 1.67 0.0167 Screens shown/Incr 3.33

0.0333

 

 

 

Screen Display 1 Screen Display 4 Screen Display 7 Screen Display 1 Screen / CPU cycle 0.0333 0.0333 0.0333 0.0333 CPU Time / cycle 0.0167 0.0167 0.0167 0.0167 0.0167 0.0167 0.0167 0.0167 Time per cycle 5.511 5.5444 5.5778 5.6112 Nr of the CPU Cycle 328 329 330 331 332 333 334

335

 

post-42078-0-28250000-1506243758_thumb.jpg

Edited by francoisadt
Link to comment
Share on other sites

  • 2 months later...

 

My own POV, for those who care. First of all, I have no stake in this at all, as I rarely play many games anymore, and concentrate on development of Stella, its debugger, etc, and above all its accuracy. So what some person got as a high score, and bragging rights, etc, are irrelevant to me. What I'm really interested in is whether the ROM that's being analyzed is the same as the one in the cart that Todd used for the alleged high score. If the ROM is different, then we need to find and analyze it. But if it's the same ROM, I wonder if there is something happening that allows the 5.51 to occur that can only happen on real hardware, and not in emulators.

 

So obviously my POV is wrt emulation, and I would be very interested to see if some undiscovered (up to this point) aspect of the real console could be influencing the results. Because from my POV, that would be something that definitely needs to be fixed in emulation.

 

And this is why Stella is brilliant because of the attention to accuracy (very important). I believe Byuu the Higan developer has been looking into random "dirty" memory on start up before var_clear aswell. Some game doesn't emulate correctly and they think it has something to do with clean memory on start up (even though it gets cleared anyway).

 

It's a very interesting subject how the memory state before being cleared can cause all sorts of results. They're even talking about different systems having variations of dirty memory down to the capacitors and how every piece of hardware (system) is slightly different from the next. All this hardware talk is on Byuu's message board/forum however you need to sign up and become a member to view. You may be interested or collect some unknown useful information that corresponds to the 2600.

 

Anyway that's why when i hear this sort of attention to detail my ears prick up. Also thanks for all the work you have done with Stella IMO there is nothing in comparison.

 

Link to comment
Share on other sites

 

Yes, but I want to know how the real hardware is differing, so that I can properly emulate it. The actual highscore is quite irrelevant to me, only that it seems to have happened yet defies logic of ROM analysis. From my POV the why is important.

There is also accusations made about the impurities in the silicon used for the chips from system to system ect. Im no guru with the hardware side of things but i know those guys really look into the randomness of hardware components to achieve the same goals you mentioned. Might be worth seeing if Byuu is up for a chat and share/compare notes as you both may benefit. Byuu is all about sharing his knowledge with other developers to gain more knowledge and produce as accurate emulation as possible.

 

Edit: I forgot to mention there was alot of talk about the crystal oscillators used having randomness slightly effecting timing depending on temp, manufacturing date and the crystal itself ect. Apparently it's hard to replicate randomness like the real hardware and most emulators usually run uniform (so i hear as for clock speed ect) . So many factors that come into play i guess.

Edited by Tony The 2600
Link to comment
Share on other sites

 

It never will be settled. Like trying to solve a murder 40 years later. There is no DNA in this case so it is all just opinions and speculation.

 

The way I see it, it's already been debunked. Simulations have proven it, humans have proven it, and shortly Ben will prove it. The only people still defending this are grasping at straws with crap like 'timing differences between models' and 'power fluctuations'. Even if something like that could allow someone to get Todd's time, they'd be considered cheating since they're hardware errors. So that means that either a. Todd is lying, or b. Todd did get the score he said he did, but it shouldn't count since it was due to a glitch. Either way, that record shouldn't stand. I really don't see any other outcome.

  • Like 1
Link to comment
Share on other sites

So that means that either a. Todd is lying, or b. Todd did get the score he said he did, but it shouldn't count since it was due to a glitch. Either way, that record shouldn't stand. I really don't see any other outcome.

It seems, for some people this is not about the score but about a. .

Edited by Thomas Jentzsch
  • Like 1
Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

Loading...
  • Recently Browsing   0 members

    • No registered users viewing this page.
×
×
  • Create New...