What Are The Limits?

3 months ago

Speedcubing Thoughts

I'm frequently asked about my thoughts on the limits of speedcubing. Are we almost there, or is there still a way to go? What times are possible on different sized cubes and other events? I've decided to explore these types of questions in this blog post, and share my thoughts and opinions.

I’ll begin with the standard 3x3 cube. A speedsolve has two determinants - move count, and turning speed. For example, a 60 move solution executed at 10 turns per second (TPS) will yield a 6 second solve, and a 40 move solution executed at 8 TPS will take 5 seconds. In my opinion, the upper limit for turning speed is probably 11-12 TPS on average with the CFOP method and its accompanying subsets and variants. This is based on both experience and observation - at the end of the day, there are physical restrictions to turning speed. Of course, it’s possible to achieve higher TPS on individual solves if the solution is rotationless, finger-friendly, or easy to execute in any other way. I’ve also assumed near-perfect lookahead in this 11-12 TPS number, which I believe is definitely possible. However, I still think there’s a lot of room for improvement in the lookahead of even the top speedcubers, particularly in early F2L pairs, and especially for the recognition of last layer cases. We're a long way off 11 TPS.

On move count, that’s an entirely different story. There also exists a massive trade-off between TPS and move count. That is, efficient solutions for solving the cube are often difficult to execute at a high speed. Additionally, looking and thinking ahead also becomes far harder as solutions become shorter, as generally these solutions require more active thinking during a solve. Ideally, a speedsolve should be entirely subconscious and automatic. CFOP has been heavily optimised since its early days, and from my perspective, most significant discoveries have already been made. It sounds a bit pessimistic, but I don’t believe there is anything that can revolutionise CFOP solve times in the way that things such as ZBLL, COLL, OLLCP, WV, VLS, and other subsets have done. Any improvements now will probably be very marginal. Brest (Rob Stuart) kindly reconstructed an average of 100 I posted a little while ago and found my average move count to be 58 STM (Slice Turn Metric) and 62 ETM (Execution Turn Metric) respectively. I certainly think this could go down to the low 50 range with more intelligent solutions and knowledge of more last layer algorithms.

For CFOP, hypothetical average TPS of 11-12 and average move counts in the low 50s suggest the potential for global averages around the 5 second mark. I believe this is theoretically possible, but the dedication required to get to that level is ridiculous. Time will tell - it might take decades for someone to actually achieve this. As speedcubing grows, the talent pool increases, and I wouldn't be surprised if the person who achieves this hasn't yet started speedcubing at all.

There is also an entirely separate debate on the advantages and disadvantages of CFOP and the Roux method. Many suggest that the Roux method has even more potential than CFOP, given its lower move count on average. Again, it comes back to the trade-off between move count and TPS - if we accept that Roux solutions are more sophisticated/complicated and potentially require more lookahead ability, then all things being equal, the TPS of those solves will likely be lower. A question I have for readers is, do you think it is realistically possible to achieve the same TPS (STM) in a Roux solve as opposed to a CFOP solve, and why or why not? In my opinion, the methods have similar overall potential - it would have been interesting to see a speedcubing community evolve in which the majority of members used the Roux method. 

Taking a quick look at the Evolution of Records page on the WCA reveals something akin to an exponential decay function, where various 3x3 rankings are charted over time. Improvement in the early years of the graph can be put down to improvements in hardware and methods, and now the gradient of the graph is gradually flattening out. The world record single has dropped only 0.93 seconds since the middle of 2011, and the world record average just over 1 second in the same time frame.

Enough waffle, it’s time to make some actual predictions!

The official single world record for 3x3 will eventually become scramble-dependent. It will remain prestigious and never get silly like the 2x2 single, but I believe that one day it will be virtually unbeatable. A single under 3 seconds in competition is possible, eventually. Even the world record average will ultimately get to the stage where top cubers need to execute an easy set of scrambles particularly well to break it. I’m not sure if we’ll ever see a sub 5 average in competition, but sub 5.5 will definitely happen, my guess is before 2020. I also predict that the WR average will be sub 6 by the end of 2018. Records are still broken extremely frequently in WCA competitions, and I think that there will come a time (in the distant future) when the focus becomes less on records, and more so on winning major competitions and titles, as is the case in many other sports/competitive hobbies which have a far longer history than speedcubing. Globally, I think a sub 5.5 average of 100 is possible way way down the track, if there are enough cubers pushing and challenging one another to achieve it. Speedcubers have always found a way to break barriers which were previously believed to be unbeatable. When I began speedcubing, a sub 1 5x5 single was considered almost impossible. Now, sub 40 is possible.

Throwing out some wild guesses for other events:

4x4: A lucky single around 16-17 seconds will eventually be achieved, and averages will peak at 21 seconds or so.

5x5: A sub 40 single is possible (perhaps imminent!), as well as averages below 45 seconds.

6x6: As the cubes get bigger, it's harder to make accurate predictions. I don't think a sub 1 single is possible on the 6x6, 1:10 might be about the limit with some luck. Low 1:20 averages are probably achievable.

7x7: Someone will get an unofficial sub 2 single in the next 12 months or so I'm pretty sure. However, I don't know if that implies sub 2 averages are also possible. 

I might write up a blog post at some other time about the evolution of hardware for big cubes, and significant milestones and events in those events, as I believe it's very interesting.

3x3 OH: I think we'll definitely see people reach sub 10 global averages, but perhaps not too much further.

Let me know what you guys think about the limits of speedcubing!


Thoughts on this blog

Notify me of updates
Dylan Teboe

Dylan Teboe Posted 3 months ago

Just for fun, I'll add my predictions.

 

2x2: We've probably already hit the limit for single if we're being honest, unless Moyu or Gan or Qiyi releases something revolutionary that Lucas or Feliks buys, then goes to a competition on the exact day that a computer algorithm will beat the odds 1 gazillion:1, and beat 0.49 with a 4 move R U R' U'. 

3x3: I hope sub 4 will be broken at some point, and after then 3x3 single will be as important, meaningful, and totally not scramble based, as 2x2.

4x4: Sub 20 in the near future, by Feliks or Max Park

5x5: Sub 40


Cuberious The SpeedCuber

Cuberious The SpeedCuber Posted 3 months ago

FELIKS! Will you do a tutorial on how you handscramble? Like how you scrambled your Gans 356 Air when you first got it?


Aneurin Hunt

Aneurin Hunt Posted 3 months ago

And in five years these predication will be laughed at. 

improvement is definitely slowing down but for the less popular events they have more room to improve. Clock and feet may still have room for improve those some people are getting really fast. 


Raghav Kulkarni

Raghav Kulkarni Posted 3 months ago


Hey Feliks can you make a complete oll sheet

 


david teh

david teh Posted 3 months ago

 Hey Feliks. What do you predict the world record will be (If someone else beats it ) in 2018.


Kurt Ramirez

Kurt Ramirez Posted 3 months ago

Feliks! te conoceré en Perú!

 


Diptoneel De

Diptoneel De Posted 3 months ago

Hi Feliks, well do you've some old videos of yours showing you solve the cube with sub 50-sub 45 averages??


Thomas Curnow

Thomas Curnow Posted 3 months ago

Well if feliks could get 3.81 in competition...


Lisa Rengers

Lisa Rengers Posted 3 months ago

meh


Lisa Rengers

Lisa Rengers Posted 3 months ago

I use CFOP (avg 16-17). I recently looked into roux and zz, because i never did before. I think it's hard to decide if one of those methods is actually better for me. I guess I'll just have to try them out... Also it's so weird that CFOP is just so evolved compared to other methods. Does that make CFOP faster? Maybe you could do a post about other methods and your experience with them..?

Well that's it. Cya Feliks :)


Yee Heng Chow

Yee Heng Chow Posted 3 months ago

I think that Feliks or Max would dominate 3x3


Charlie van Ooran

Charlie van Ooran Posted 3 months ago

hey feliks, do you think eventually everyones SD cards will run out and the WCA will fall into chaos and anarchy due to the lack of evidence of records?


Charlie van Ooran

Charlie van Ooran Posted 3 months ago

sorry serious question now, so with 4x4 how people began using yau to achieve faster solves than redux 4x4 solves, do you think that that'll eventually happen similarly with 5x5? where you among other top cubers will eventually swap to yau and redux will seem inferior? I'm probably going on a big tangent with this as yau5 is like very different to yau on 4x4, but I was just wondering : )


David Rostek

David Rostek Posted 3 months ago

 Do you think is it possible to find a better method than CFOP?


Corner Cutter

Corner Cutter Posted 3 months ago

Interesting article!

What do you think about the limits for the 2x2 average and single?  

And what are your PB's at home for all the events 2-7?


Feruz Gulomov

Feruz Gulomov Posted 3 months ago

 Hi Feliks! Just wanted to ask, is it possible to get sub 30 ao 5 on megaminx? 

P.S. I think that YOU will be the first speedcuber, which will do sub 6 ao 5 on 3x3 in WCA. 


Cedric R.

Cedric R. Posted 3 months ago

Hi community and hi Feliks!

Very nice article and interesting thougts!

In my opinion there arent any marks or maximal times to reach.

I know, it sounds unrealistic, but I think, in distant future the 3x3 record will be sub 2 seconds, maybe 1.

The cubing hardware ist still frequently improving and there are many upcoming functions which can improve the times.

One example of this is the magnetic function of the gan 356 air ultimate magnetic.

Further more, the speedcubers are becoming better and better by improving the methods frequently.

One day there will be a "master-method", which will be used by every top cuber and will enable cubers to reach move counts round about 30.

The only problem of speedcubing is, that professionel speedcubers cant put so much effort in cubing like professionell guys from other sports can,

because a speedcuber will never earn enough money with cubing.

Even though I think that sub 2 seconds solves will be possible in distant future.

 

Faithfully, Cedric


Samuel Lai

Samuel Lai Posted 3 months ago

As far as hardware limits go, I think a lot can be done for 6x6. Currently I can name a possible problem with every 6x6 out there, including the "big three". Spring swapped YuXin barely corner cuts, WeiShi is big and quite slow (Spring swapped WeiShi is awesome, I used KungFu 4x4 springs in mine), WuHua feels a bit tight and I can get pops on it. I have some ideas and have seen some 7x7 mechanisms that have yet to be used on 6x6, and are definitely possible to convert to even-layered cube. In comparison, the 5x5s, 7x7s and 8x8s are quite ahead of the 6x6s. (There are only 2 8x8s really, but ShengShou 8 was well received in its time, better than the ShengShou 6, and YuXin 8 has been said to be the "Best of the HuangLongs")

I also think there are some areas that should be tried in 4x4, like making 61mm or 59mm cubes. Maybe more people may find the exact size they need.

I know hardware better than I know methods, so hardware side is the most I can predict.


daniel evand

daniel evand Posted 3 months ago

I have noticed that world records generally have half cross +F2L and half OLL+PLL so if someone could get a sub-2 cross+F2L in a solve and a sub3 regularly then if they learn 1LLL then it is definitely possible for sub 5 averages however it requires near-instant recognition out of 3668 cases which would not be easy but possible. you can sub-1 any alg I assume feliks so if you can sub-1 every 1LLL algorithim then it would mean very fast solves.

so my predictions for 2025:

3x3x3:  3.99/5.32

4x4x4: 16.47/21.24

5x5x5: 36.89/41.72

7x7x7: 1:55.78/2:15.05

pyraminx: 0.99/1.78

skewb: 0.99/1.78


Purva Parmar

Purva Parmar Posted 3 months ago

Maybe we would need more mathematical research into the different cubes.

Any mathematician friends out there??

 

There are some universities which indeed do this. Maybe we can perform simulations on a computer, to find an estimate of limits of speed cubing.

 

To check all the staggering number of permutations, we would definitely need a supercomputer. But it takes even a supercomputer quite some time to check all the cases for the 3*3, forget the larger cubes.

 

A Quantum Computer can solve this problem.

Any scientists into Quantum Computing out there?

My Idea:

We need data about different types of hands, fingers, their shapes and sizes.

Then we need data about the thinking capabilities about different types of brain structures.

We need complete data about the areas of brain which are into pattern recognition and processing of information.

 

We then need a nice info of the various cubes, which is the only thing (I think) we completely have.

 

We then need data about how our fingers exactly behave when they turn the cube.

 

We need more data about parameters I didn't mention here. My cuber friends, give us more parameters.

 

 

Of course, these factors are going to be unique for all the people.

 

We can then incorporate all these parameters into a nice, decent and efficient computer program. Then we can run the simulations.

 

Nothing can beat a Quantum Computer, but as we don't have that, we can think about cooperation.

We can combine all of our computers' power to run the simulations. That is, each person participating in the programme​ runs a given number of simulations.

 

This, if undertaken, would be a massive project.

We would need mathematicians, biologists, neurologists, computer scientists, expert programmers, expert cubers, ..., and participation of a lot of common people. We would even need funding for the project.

 

We can then even look out to different methods of speedsolving. 

 

 

I'm sure there are cubers in almost all the mentioned professions.

 

If undertaken, I would be one of the first to participate.

 

Come on, who's going to join me? Voice your opinion.

 


Purva Parmar

Purva Parmar Posted 3 months ago

I would like to add something.

 

Many cubers like to just watch how the pieces move, and they experiment different things.

Many are into (unofficial) research into cubes.

 

We need a unique and universal platform to publish the discoveries. It's different from cubing communities, in a sense that it focuses on the discoveries in different methods of speedsolving.

 

Of course, we can have multiple local platforms as well, which would organise their work with a main authority.

As WCA has been the official authority for regulation of competitions, we can have a unique authority into this matter.

My cuber friends, I hope that we do start such a platform, so that even the smallest discoveries can be studied by the experienced cubers.

It's like scientific research. One person discovers something, other builds upon it. History has been a witness that some of the biggest and most important principles of science were built on some basic ideas discovered by less-known people.

Any researchers out there? Can you help us with the correct approach of publishing a discovery?

Feliks (Sir), since you're one of the top cubers, whatever you do has a great influence. You might be having acquaintances around the world that might make this thing possible.

 

Now, not only Feliks, we all have the responsibility of building upon the existing ideas, and develop new ones.

 


shubham Kumar

shubham Kumar Posted 3 months ago

Which 4×4 cube is better Qiyi Qiyuan, MoFang JaioShi MF4, Qiyi Wind Cloud, YJ yusu, KungFu CangFeng, KungFu JoQue? Can any one help me?


Aaditya Sikder

Aaditya Sikder Posted 3 months ago

I think sub 3 is possible.Just kidding

 


Tien Tran

Tien Tran Posted 3 months ago

I'm pretty sure sub 6 or even sub5 averages are definitely possible in the distant future. I think that sometime later, speed cubers will do an easier way to do last layer, like using 1LLL, but changing the orientation and the permutation of the last layer by doing f2l in a certain way so that you get to know a reasonable amount of the cases instead of nearly 4000 of them. Maybe a hundred years from now, Speedcubers would make a method that's 30 moves or less. Anything could happen.

 

 

p.s. You forgot 2x2


Tien Tran

Tien Tran Posted 3 months ago

the ultimate method that will be created in the future will be called God's Method.


Tien Tran

Tien Tran Posted 3 months ago

SHUBHAM KUMAR easy its the Qiyi Wuque


Kaedan Hitt

Kaedan Hitt Posted 3 months ago

I think it is possible to get sub 3 I have got a couple last layer skips you could get a x cross and last layer skip could get you sub 4 and maybe sub 3 if you get easy pairs


Feliks Zemdegs

Feliks Zemdegs from CubeSkills Posted 3 months ago

Charlie - one thing I would like to do after this WC is to practice Yau on 5x5 a lot more and see how I go. I think the method has similar, if not more potential than standard reduction.

David - CFOP is just the starting point, there are many many enhancements to the basic method. Roux can certainly rival it in terms of speed, but I'm not sure we'll see something completely different which is much better than either of those methods.

Feruz - Not sure, ask Juan Pablo! I think yes, but it's probably on the edge.

Samuel - I certainly agree, my 6x6 is very tight and I think better hardware could easily drop my times some more. 

Daniel - cheers for the predictions! The nxn ones look quite good, perhaps the 7x7 average is a bit conservative though :)

Purva - I think this sort of thing naturally occurs within the speedcubing community, it's hard to see some sort of official research group being formed :p

Shubham - I've heard good things about the Qiyi and the KungFu

Tien - Interesting prediction! That sort of thing would require a ton of dedication, are you up for it? :p


Feruz Gulomov

Feruz Gulomov Posted 3 months ago

Feliks, can YOU shoot a video on a subject: 'How to use ispection time' ? 

 


John Lewis

John Lewis Posted 3 months ago

 

I have a few things to add.

I have been about neck-deep in method development for about a year now, and here are my thoughts on the limits for certain methods:  

ZZ: I think that this has a little less potential that CFOP for two hands. However, I think this method is especially suited for one-handed solving due to the 3-gen nature, rotationlessness, and relative efficiency. I feel if enough people switch to this method, OH world records will begin to be set with it.  There are some variants of ZZ that are emerging that give you some incredible advantages, such as 1LLL's under 200 algs, and other stuff.  In my opinion the 4 fastest variants are in order ZZ-b, ZZ-VH(COLL-EPLL), ZZ-WV(tied with ZZ-VH), and ZZ-CT. ZZ-a is not on here because there are too many algorithms and the fastest ZZ-a solver is much slower than the fastest ZZ-b solver(Tao yu)

CFOP: This may be controversial, but I think CFOP with only one color has less potential than Roux.  However, with either bicolor or full Color neutrality I think it will achieve WR singles far into the future.  CFOP has a higher standard deviation than Roux, so fast singles are more likely.  LL algs seem like a good Idea in theory, but remember that the WR single and average were both with OLL-PLL. Also it will be the method of choice for big cubers for a while back 

Roux: In my opinion, a perfect world where 50% of the people in the world used Roux and 50% used CFOP, WR single would be with CFOP and WR average with Roux. This is due to some similar trends I have observed as well as my own experience.  Roux generally has a much higher consistency than CFOP, which is shown in this example: Kian and Feliks have similar averages, however Kian can count the amout of 4's on both his fingers when Feliks gets them very often.   People Like Alexander Lau, Kian Mansour, and Kavin tangtartharakul have shown that 

so in short, in my opinion the decision between Roux, CFOP, and ZZ should be up to personal preference, as they really don't limit your times all that much.


shahzad sheza

shahzad sheza Posted 3 months ago

Feliks iam new to this website.I have a question that is the cube you use really matter? please help me

 


shahzad sheza

shahzad sheza Posted 3 months ago

Iam not good at cubing  but I can solve it in 30 to 40 seconds,how can I increase my speed?

 


Jj Jj

Jj Jj Posted 3 months ago

TIEN TRAN 

 


Jj Jj

Jj Jj Posted 3 months ago

Tien Tran 

There will be a sub 4  because in reality humans can be fast about it. Like this guy in my bus . You should watch his youtube channel; Tien Cubes


Cuberious The SpeedCuber

Cuberious The SpeedCuber Posted 3 months ago

Can a bunch of different methods combine to make one ultimate method?

Also, can you reach 13.4 or 12.35 TPS on any SpeedCube?


Riley Sage

Riley Sage Posted 3 months ago

On the efficiency side of the board, I have an idea that might pay out well, but will definitely take a lot of time and effort. My idea is to have a learning program (I don't remember the actual term for it) analyze a billion or so optimal solves from random WCA scrambles, and then have it perform further training via virtual cubes. Then someone could analyze these solves and try to recreate this kind of 'mindset' that the program has adopted and base their solves on that. I don't think it would create an entire method, but it could shine some light on alternate routes for solving.


Aaditya Sikder

Aaditya Sikder Posted 3 months ago

Hey,Feliks are you good in maths?

 


Dylan R

Dylan R Posted 3 months ago

That's pretty michy


Aurora Pratt

Aurora Pratt Posted 2 months ago

I think Feliks once made a Sub4 ina video


Ethan Jan

Ethan Jan Posted 2 months ago

Do you think there will ever be a sub-1 average in 2x2?


Nguyen Vu Hoai Nhan

Nguyen Vu Hoai Nhan Posted 2 months ago

Hi Feliks, do you think a sub-30 in megaminx is possible?

 


Simon Geertsema

Simon Geertsema Posted 2 months ago

Lucas Etter has a sub 1 ao5 on 2x2 At home


Diptoneel De

Diptoneel De Posted 2 months ago

Hey Feliks, do think sub 1(hour)17x17 is possible????


戴 卓

戴 卓 Posted 2 months ago

6666

 

 

 

 


欣惠 刘

欣惠 刘 Posted 2 months ago

l agree with you 戴卓


Jeffrey VanderKuyl 2

Jeffrey VanderKuyl 2 Posted 2 months ago

  1. Probably only Feliks himself could get sub 1 hour on a 17x17 Diptoneel De

Gurustat  Singh

Gurustat Singh Posted 1 month ago

Hey Feliks do you think a sub 28 in megaminx possible  , And what are your predictions on 2×2 WR , it is tough to break as it is in milliseconds 0.49 


Jeffrey VanderKuyl 2

Jeffrey VanderKuyl 2 Posted 1 month ago

Hey Feliks do you got to any comps in the USA and if you do where?


Alan Jia

Alan Jia Posted 1 month ago

'I also predict that the world record average will be sub-6 by the end of 2018." Or the middle of 2017.


Byron Shan Belmonte

Byron Shan Belmonte Posted 1 month ago

feliks do you think you can still beat the current 3x3 WR


Tieg Selfors

Tieg Selfors Posted 3 weeks ago

Feliks.  DO you think sub 4 3x3 will happen by 2020?


Jonathan Liu

Jonathan Liu Posted 1 week ago

Sub-6 3x3 avg by the end of 2018? More like June 2017! :D


Charlie Harrison

Charlie Harrison Posted 5 days ago

When do you think cubing will die out completely?



Become a free member to post a comment about this blog.