ChessBase Logo Shop Link
Language :
Search :
OK

Komodo 8: the smartphone vs desktop challenge

by Albert Silver
9/24/2014 – After hearing two strong players argue that the only real progress in chess engines in the last ten years was due to faster computers a special match was played to challenge this idea. Komodo 8 ran on a smartphone while a top engine of 2006 used a modern i7 computer that runs 50 times faster. This is the difference between Usain Bolt and the Concorde. Guess what happened?
 

Over the last 20 years, and even more so the last decade, we have seen technology advancing by leaps and bounds in ways that would shock someone from only 1990. Hardware has advanced so much that our smartphones of today (the first iPhone dates back only seven years) are far more computer than phone, but software has also advanced hand-in-hand with this process. The strength of chess engines has also taken off, but it is sometimes hard for the layman to differentiate how much of the progress is due to hardware and how much to software. The answer is both of course. 

Still, it was a bit of a shock to listen to a conversation between two strong players, one of whom is a grandmaster, claim that engines had really not progressed at all in the past ten years, and that the Elo leaps the programmers were taking credit for, were entirely due to the faster hardware. Needless to say, for someone who knows several of these programmers, and the enormous time and effort expended, to see them classified as con artists did not sit well with me.

I realized the statements were not meant maliciously, and were sincere in their utter ignorance. After some thought I realized the vast majority of users are probably unclear on just how much progress has been made in pure programming terms. There are of course ratings lists of engines, but the numbers are so outrageously high over human ratings that it becomes hard to fit them into any normal perspective. Consider one of the top ratings lists over the past years: CCRL (Computer Chess Ratings List).

Komodo 8 tops the ratings lists for computer chess

At the top on a single core is Komodo 8 with a mere 3232 Elo. For most of us, the near 2900 rating of Magnus Carlsen is already approaching science fiction, so where does 3232 fit in the picture? That is today though. The best engine 10 years ago was in fact Shredder 8, but since the list does not go back that far, we will make do with Shredder 9, rated…. 2725.

Although a top engine in its day, the lists suggests that in pure software there is a 500 Elo gap

That suggests that software has advanced over 500 Elo in ten years, without factoring in the faster computers themselves. Can this be true?

I decided to run a small experiment. Since Komodo 8 exists on both the desktop as well as the Android smartphone, I decided to play a small match between it on a smartphone, facing Shredder on a modern, top of the line quad-core i7 processor. I ran a small test to compare the speeds of the platforms, by running Komodo on the start position and seeing how long it took to reach 20 plies. The result was that it ran about 50 times slower on a smartphone. Not 50%, no, 50 times.

To put that in perspective, if Komodo were racing at the speed of Usain Bolt at his fastest (roughly 45km/h) then Shredder would be racing in a Concorde at 2250 km/h. Therefore if Komodo’s edge over the best of yesteryear was in fact only 150-200 Elo, then the fantastic hardware advantage by Shredder should allow it to crush it. Although I actually own Shredder 8, and could dig it up, it is unable to make use of more than one core (2004 after all), so instead I used Deep Shredder 10, from 2006, which is not only much stronger but able to make full use of the quad-core i7.

A view of Shedder 10 using an i7, facing Komodo 8 running on a smartphone

Three positions were chosen to be played with reverse colors, each engine playing it both as white and black, at a time control of fifteen minutes with a ten second increment. Why only six games? Quite simply because I had to do this by hand, and moreover, was forced to manually tell Komodo when to play. The reason is that it was not designed to play against an opponent on a smartphone, just as an analysis engine, so I would enter Shredder’s move, let Komodo analyze, and then tell it to play its move.

 

I took the opportunity of recording the sixth and last game in such a way that readers might
follow the analysis of Shredder 10, but also Komodo 8 on the smartphone at the same time.

The final result after six games was a merciless crush by Komodo 8 with four wins and two draw, winning by 5-1. Had the match been closer, I might have considered extending it to see how it progressed, but 5-1 is pretty clear. As a reference, the performance over the six games was a 280 Elo lead, therefore if you accept that Deep Shredder 10 on an i7 is at least 2800 Elo (it was rated 2770 on a slower single-core system), then Komodo 8 is well over 3000 Elo on an Android smartphone. For the record, the smartphone in question was an LG Optimus G Pro from 2013.

The games of the match

Deep Shredder 10 UCI–Komodo 8, smartphone½–½D51Rapid 15m+10s1Rio de Janeiro, Brazil22.09.2014
1.d4 f6 2.c4 e6 This was the starting position. After this, the engines were on their own. 3.c3 d5 4.g5 bd7 5.e3 h6 6.xf6 xf6 7.f3 c6 8.e2 dxc4 9.xc4 d6 10.e4 e7 11.0-0 0-0 12.d3 a5 13.e5 d5 14.b3 b6 15.xd5 cxd5 16.c2 g6 17.a4 d7 18.b3 a4 19.c2 c8 20.ac1 g7 21.e3 b5 22.fe1 d7 23.a3 c7 24.b1 fc8 25.xc7 xc7 26.h4 c4 27.d2 a6 28.g3 b5 29.g2 c6 30.h2 h5 31.f3 c7 32.c1 xc1 33.xc1 e2 34.c2 b5 35.d2 b4 36.xe2 xe2 37.axb4 b5 38.d2 xb4 39.b1 a5 40.f3 f6 41.exf6+ xf6 42.e3 e5 43.dxe5+ xe5 44.xg6 b6+ 45.d2 xf2 46.c3 d7 47.e2 d4 48.xh5 e3+ 49.c2 f5+ 50.d1 f2 51.d2 d3 52.g4 h7 53.c1 d4 54.xd3 xd3 55.e8 xh4 56.xa4 c4 57.c2 g5+ 58.d1 b5 59.f5 a4+ 60.c2 e8 61.f5 c6 62.e6 a4+ 63.e2 ½–½

If anyone had any doubt as to the genuine progress of chess engines, and the merit of buying new versions, they can lay these concerns to rest. Chess engines have not only progressed over the last ten years, but there is reason to believe they have outpaced the hardware even. If you want the latest and greatest, Komodo 8 is the king of the hill.

Komodo 8 can be purchased in the ChessBase Shop

Born in the US, he grew up in Paris, France, where he completed his Baccalaureat, and after college moved to Rio de Janeiro, Brazil. He was champion of Rio de Janeiro with a peak rating of 2240, and was a key designer of Chess Assistant 6. In 2010 he joined the ChessBase family as an editor and writer at ChessBase News.
Feedback and mail to our news service Please use this account if you want to contribute to or comment on our news page service

See also

Discuss

Rules for reader comments
    Not registered yet? Register
DELTAMAX2020 DELTAMAX2020 9/24/2014 10:37
Openin book is also an important factor.
DELTAMAX2020 DELTAMAX2020 9/24/2014 10:40
According to my own calculations when the CPU power doubles it has 100 elo plus for every engine.
Werewolf Werewolf 9/24/2014 10:50
I ran a match recently between Stockfish 5 on the iPhone 5s and Deep Fritz 10.1 running on a 16 core workstation. The result was similar - Fritz lost almost every game.
DELTAMAX2020 DELTAMAX2020 9/24/2014 11:14
ELO calculation is based on theaverage ELO of oppenents.For example if you have 100 diferent engines on the ELO list like 1st 2800 2nd 2700 ...etc it does not matter when you run the same test on a 1 millon times faster computers. ELO ratings in this case stays nearly same for all the engines. 1st engine 2800 2nd 2700 etc.But when you make comparison outside the same list you can see that on a one million times faster computers, power of the engines are skyrocketed compare to human ELO or "real" ELO.This is because of the ELO formula.
Gurcan Uckardes Gurcan Uckardes 9/24/2014 12:50
Well done for better pushing Komodo sales. I'm not against that as a mobile chess freak. According to the article, it's fine but there's a detail i don't find accurate. This 50X ratio of power doesn't make sense unless the progs are the same. If Komodo mobile plays vs Komodo PC it's OK but here in this experiment, no! However it's good to see the proof that modile CPU's are not toys anymore and some of them are really challenging. Regarding the calc power vs ELO, exact versions today can reach ++50 to 100 ELO with double time but it does not apply to other engines directly. Another factor which breaks this rule is the depth vs time. Every engine has its own curve and is optimized for a range of hardware power. We can't assume Komodo would reach 3600 ELO just by using 64 cores (though it's not possible today). The ELO gain is less and less per extra depth beyond a threshold.
PurpleUnicorns PurpleUnicorns 9/24/2014 03:01
Opening book isn't really that important. Top engines usually don't make large mistakes with or without book anyhow.

I've read from credible sources that a doubling in CPU power is around 50-70 elo, however it is less when the increased power comes from an increased number of cores.

The improvement of engine software is ridiculously massive over the passed few years. Significantly larger then any CPU improvements in the same time-frame. Some people probably might think Rybka, Fritz or shredder are top engines nowadays, but those days are long passed. Their newest versions are only the #7, #14 and #20 respectively according to CCRL. Top engines today are Komodo, Stockfish and Houdini in that order. Those three engines will tear any older engine to pieces with their bare hands.

The problem is people have trouble imagining just how ridiculously powerful these engines are, which makes it even harder for people to comprehend that the engines are continuing in leaps and bounds. This to a large degree is probably due to the #2 engine Stockfish which is open source and can get downloaded for free after a trivial google search.

That being said, I believe many will prefer Komodo 8 style when analysing over Stockfishs. In any case I think it's completely legitimate to support these brilliant programmers.
izzat1987 izzat1987 9/24/2014 03:08
It's really funny this Komodo campaign propaganda ,for the sake of the argument ,if we agree that komodo is the strongest engines ,we are not talking about a huge gap between it and other free and open sourced "lower" rated engines, which will not make a difference in the eye of a 300-500 less rated player !
Bertman Bertman 9/24/2014 03:55
@Deltamax2020 Actually, engines gain about 50-60 Elo per doubling in speed. This is easily verifiable in engine ratings lists that test single-core, dual core, and more, and has been the reality since the advent of multi-core engines.

@Gurcan The 50x comparison between Komodo on a smartphone and desktop is valid since the point is to show that Komodo is giving up a 50x speed edge. If it also ran on a desktop against Shredder its speedup would be.... 50x.
MarriedRhombus MarriedRhombus 9/24/2014 04:13
These engines are the main tools of the cheaters.
thlai80 thlai80 9/24/2014 05:06
A lot of details were left out however. Which smartphone with what processor, and what is the depth when the move was forced for Komodo on the phone? Then, there's the setting of Shredder on desktop, the contempt value, the memory used, is it tournament book, etc. Finally what exactly is top line quad core i7 processor (1st gen, 2nd gen or 3rd gen) what is the amount of ram, not to mention OS (eg. windows version since it is Shredder from 2006) etc? In terms of calculation speed, what exactly is the position or nodes per sec for both setups? For pure computer strength comparison, I would take out opening book altogether!
Bertman Bertman 9/24/2014 05:23
@thlai80 Most of your questions are answered in the article and video. Bear in mind that the generation of i7 used, the model of the smartphone, are all irrelevant, since what really matters ultimately is to know how much of a speed difference there was between the two during the test. That said, for the sake of disclosure, it was an Ivy Bridge i7, and the phone is an LG Optimus G Pro. Regarding NPS, the answer is there to see. Note that you cannot compare the NPS of two completely different engines.
alfilferoz alfilferoz 9/24/2014 06:20
It would be interesting to repeat the experiment with Komodo 8 in the same i7 desktop computer against Komodo 8 smartphone.....
iSeeThis iSeeThis 9/24/2014 06:30
It should be noted that Deep Blue was one of the most powerful supercomputers in the world in 1997, running a massively parallel RISC System/6000 with 30 nodes each containing a 120 MHz Power2 Super Chip microprocessor. Someone estimated that Deep Blue was around 2800 then (well, that's enough to beat the then #1 chess player on this planet; Kasparov!) Roughly, Shredder10, also 2800, is supposed to beat Deep Blue 50-50. Guys, it means that an app on smartphone today can easily beat Deep Blue like father beats his child!
GalacticKing GalacticKing 9/24/2014 06:31
Nice work, Albert! I love running engine matches, so kudos for taking the time to do this. So far, Komodo 8 has been the only worthy challenger to Stockfish 5 on my android phone!
NeoNietzschean NeoNietzschean 9/24/2014 08:33
Yeah, I too have heard people claim that engines are only better now because of hardware. It's obviously false to anyone remotely familiar with engines, but it's nice to have a highly publicized test like this.

It is a shame that it couldn't be automated, as the 6 game limit is probably what people who want to dismiss the result will point to, since 6 games is such a small sample.

Of course the result would hold over a large number of games, but the doubters will continue to doubt until such a test is made public.

Also, the numbers for rating gain per doubling are not fixed at any value. The longer the time control/the faster the processors, the less gain per doubling.

With a given branching factor, a doubling in speed will buy you a fixed increase in average search depth per move, and the data bear out the intuition that buying an extra ply of depth at 10 ply is worth a lot more rating than buying an extra ply of depth at 30 ply.

Several tests at talkchess.com have indicated that at classical time controls, the gain could be as low as 30 points for a doubling in speed, even (but don't seem to ever flatten out much beyond that, interestingly.)
PurpleUnicorns PurpleUnicorns 9/24/2014 09:23
Perhaps an easy and interesting test would be to have both engines on the same hardware, but give handicap Komodo's time by factor 50 or something similar. This would make it easier to compare as well as remove the automate move transmission.
MCris MCris 9/24/2014 10:25
I don't know if anybody saw all the games -there is a drop-down menu- but when Shredder lost with White it was put to play inferior openings like the English (1.c4 e5) and an Italian variation where one pawn is thrown away in the opening (!)
Bertman Bertman 9/24/2014 10:43
@ISeeThis Does that really surprise you? 30 x 120 MHz is... 3.6GHz, which is already slower than my desktop. Plus it was split up in 30 CPUs, which means a huge loss in efficiency. Between the hardware advances and the huge software advances, yes, Komodo on a smartphone would mop the room with 1997's Deep Blue. Besides, DB did not go nearly as deeply as you might think. The average depth of the middlegame was 12 plies or so.
jhoravi jhoravi 9/25/2014 01:19
Komodo 8 in a Smartphone will destory IMB Deep Blue 2 more convincingly.
Badir Badir 9/25/2014 02:22
First a modern cell phone cpu is not 50 times slower than an i7. You can run several test to compare and most put the speed difference from 8 to 10 times faster and NOT 50. They are getting closer in terms of speed every year also.

Secondly you can't force openings on the programs as they are optimized for their library they have. With an active library they would probably dump both programs into roughly equal 20 move midgames instead of what they were forced into.

Thirdly the only test of interest would be tournament style length game say something like 40 moves per 2hours.

The article seems to be about pushing software on people instead of giving any real information.
NeoNietzschean NeoNietzschean 9/25/2014 02:32
@Badir: No, the speedup cited in the article is accurate.

What matters here is how quickly the program performs a given task on the phone relative to performing the same task on the desktop.

For chess, that difference is very, very large. For example, I've run multiple tests on my systems (phone is Samsung Mega 6.3, desktop uses an i5-3550), and it takes on average 41x as much time to reach the same depth on my phone as on my desktop.

I'm sure there are specific benchmarks where phone processors are much closer to desktop processors, but it doesn't matter for this article whether video rendering, flash processing, or compression only take 8-10 times longer on phone processors.

What matters is how much faster chess engines are, and the number cited in the article is consistent with what I've seen on my own hardware.

Just my $.02
MrL2014 MrL2014 9/25/2014 02:57
The question is if a 2006 software can make use of a 2014 processor in an efficient way. Not to say that there was not a tremendous progress on the engine itself.
iSeeThis iSeeThis 9/25/2014 03:48
@Bertman @jhoravi
Deep Blue was my Sci-fi fantasy. I then always dreamed of having Deep Blue in my palm. Just didn't dream that it would happen before I die (or even old :-D)
Bertman Bertman 9/25/2014 05:34
@ISeeThis Yes, same, but I had never really thought about it until you brought it up. In case you are curious, here are the computer logs of Deep Blue showing its analysis for each move.

https://www.research.ibm.com/deepblue/watch/html/c.shtml
joe99 joe99 9/25/2014 07:17
it is about the quality of the chess engine and not the quantity , two heads are not better than one and speed calculation shouldn't make a big factor either. if you have an Engine that knows how to play with high quality moves it will easily win against an engine that is calculating at high speed but is making mistakes deep in its calculation that cannot come to the right conclusion of a position . you could be have the fastest car in the world but if the driver can't drive then it is as useful as a bicycle with one wheel.
Rama Rama 9/25/2014 08:00
Komodo's place on the CCRL rating list was based upon opponents rated lower than Stockfish's were and on hundreds of fewer games. Why don't they just do a quadruple round robin between only the top three engines to settle this thing once and for all?
PurpleUnicorns PurpleUnicorns 9/25/2014 10:50
To those criticizing the openings keep in mind the openings were symmetric in the sense each engine played each side once so any claim of inferiority (which is ridiculous anyhow as the openings are just normal and fine choices) would if at all only bias the result towards a 50% score.

Secondly no, I'm pretty sure the engines are not optimized towards using there own opening books. If anything the optimization would be to make a book to an engines strength. Either way a more modern engine would if anything again have an edge here, though I don't believe it would make any difference at all.

Finally I can point you all to the TCEC, the Thorensen Chess Engine Competition which is probably the most respected chess modern chess engine competition nowadays. One of the stages this season will be played without book and previously there was a chess960 competition. The chess960 competition only further showed that with or without book, the top engines dominate the competition.

Speaking of that chess960 competition, Stockfish scored 25.0/28 which is ridiculous. For comparison, the newest version of Rybka scored 14.5/28 and the newest version of shredder scored 7.5/28

The engines of 10 years ago are so much weaker then today's beasts its not even funny.
PurpleUnicorns PurpleUnicorns 9/25/2014 10:52
Chessbase needs to add the feature to be able to edit posts, I made so many grammar mistakes just now =(
EddieFreak EddieFreak 9/26/2014 12:11
I agree with MrL2014. I would have liked to have seen this tested using older hardware. example PIII 600mhz. Take away all those little instructions sets that the older engine can't use.
1
© 2013 ChessBase GmbH | Osterbekstraße 90a | 22083 Hamburg | Germany |  Imprint  | Contact  | Home