Fritz 3 ELO on different computers
Moritz Berger
following conclusions concerning the playing strength of Fritz 3 on
different platforms (486er, Pentium etc.):
1. Tripling the available computing power approximately increases the
programs playing strength by 100 ELO points.
2. Tripling the time limit for the game decreases playing strength
by 100 ELO points.
Here are some ELO numbers for my computer:
| 486/50 with "turbo" | 486/50 without "turbo"
---------------------------------------------------------------------
5 min./game | 2440 | 2340
10 min./game | 2392 | 2292
15 min./game | 2357 | 2260
30 min./game | 2288 | 2185
40 moves/120 min. | 2057 | 1954
CPU benchmarks with Norton sysinfo 7.0 and PC Tools 9.0 si showed
the non-"turbo" mode to be approx. 1/3 as fast as "turbo" mode.
On a Pentium 90 the program should (optimistically) have about 2540
int. FIDE ELO when playing 5 min. Blitz. On the other hand, in regular
games (40 moves/120 minutes) playing strength should be about 2157 on
the same machine.
I should say that the program doesn't take into account the size of hash
tables, although it shows a warning that you won't obtain full playing
strength when you start it with only 64KB hash tables (standard). It also
neglects more subtle details of hardware speed such as memory speed and
size and speed of the 2nd level cache.
Since I played 60 games in 5 min. Blitz CM 4000 vs. Fritz 3, I tend to
believe that the programs evaluation of its own playing strength should
be a good hint of its "real" strength, maybe with a correction of -50
ELO. This evaluation is of course true only with todays hardware (up to
Pentium), so nobody should try to calculate with the above rules of
thumb what hardware speeds are necessary to obtain a 3000 ELO with
Fritz 3! The same is valid for the time constraints (playing strength
with 1 second/game ...).
Generally, the effect of more computation power on plaing strength
decreases up to a point where it becomes insignificant if you double it
or even use 100 times as much. This is due to the exponential growth of
the tree of possible moves. This is also the reason why Fritz 3
seemingly plays worse in non-Blitz games: The program doesn't get much
stronger with the extra time given while it's human opponents play much
better if they're given enough time since humans don't try to play chess
by using computation trees but with "true" intelligence ;-) (this is of
couse an oversimplification of the problem ...)
As always, ELO ratings are international FIDE, for American USCF ratings
add 100.
If you have trouble finding the evaluation function: It's only
available if you play with fixed time limits and in "real
tournament"-mode. This mode can be invoked by pressing the p key.
I'd like to learn some "real" results from people who actually have
Pentiums (or slower computers) and see if my approximation holds.
------------------------------------------------------------------------
Moritz Berger uzs...@ibm.rhrz.uni-bonn.de
------------------------------------------------------------------------
Mads Brevik
> Based on the ELO evaluation of the program itself, I came to the
> following conclusions concerning the playing strength of Fritz 3 on
> different platforms (486er, Pentium etc.):
>
> 1. Tripling the available computing power approximately increases the
> programs playing strength by 100 ELO points.
>
> 2. Tripling the time limit for the game decreases playing strength
> by 100 ELO points.
>
> Here are some ELO numbers for my computer:
>
> | 486/50 with "turbo" | 486/50 without "turbo"
> ---------------------------------------------------------------------
> 5 min./game | 2440 | 2340
> 10 min./game | 2392 | 2292
> 15 min./game | 2357 | 2260
> 30 min./game | 2288 | 2185
> 40 moves/120 min. | 2057 | 1954
Are these ratings also based on the built-in strength rater?
The blitz-rating (5 min/games) seems a bit low compared to
the rating of Fritz 2 (486/50-66) on SSDF's computer blitz-list.
I would expect that Fritz 3 plays at least 50 points stronger
than its predecessor at blitz-level. But I understand that
the ratings above are only estimations, so it seems we have to wait and
see.
The tournament rating though, is far too low in my opinion. Fritz 2 is
currently rated at 2.150 ELO on a 486/33-system (SSDF), or slightly above
2.200 on a 486/50-66 (my estimation). My guess is that Fritz 3 should
perform at around 2.250 ELO (at least) on a 486/66.
>
> CPU benchmarks with Norton sysinfo 7.0 and PC Tools 9.0 si showed
> the non-"turbo" mode to be approx. 1/3 as fast as "turbo" mode.
>
> On a Pentium 90 the program should (optimistically) have about 2540
> int. FIDE ELO when playing 5 min. Blitz.
I don't think 2.540 ELO is optimistically for Fritz 3 running on a
Pentium 90. My guess is 2.550 on a 486/66 and 2.650 on a Pentium 90.
Genius 2.0 (currently the strongest blitz-program rated by SSDF)
has a blitz rating of 2.630 ELO (!) on a 486/66.
> On the other hand, in regular
> games (40 moves/120 minutes) playing strength should be about 2157 on
> the same machine.
>
If you add 200 points I agree.. :)
> I should say that the program doesn't take into account the size of hash
> tables, although it shows a warning that you won't obtain full playing
> strength when you start it with only 64KB hash tables (standard). It also
> neglects more subtle details of hardware speed such as memory speed and
> size and speed of the 2nd level cache.
>
> Since I played 60 games in 5 min. Blitz CM 4000 vs. Fritz 3, I tend to
CM 4000 and Fritz 2 play almost at the same level in blitz. Fritz 2 is
perhaps i bit sharper. If I don't remember wrong, CM 4K has about 2.450 ELO
on a 486/66.
> believe that the programs evaluation of its own playing strength should
> be a good hint of its "real" strength, maybe with a correction of -50
> ELO. This evaluation is of course true only with todays hardware (up to
> Pentium), so nobody should try to calculate with the above rules of
> thumb what hardware speeds are necessary to obtain a 3000 ELO with
> Fritz 3! The same is valid for the time constraints (playing strength
> with 1 second/game ...).
>
> Generally, the effect of more computation power on plaing strength
> decreases up to a point where it becomes insignificant if you double it
> or even use 100 times as much.
Yes, unfortunately. But the increase in strength when you
double the speed (computational power) for example, varies somewhat
depending on the nature of the program. Strong programs tend to
improve less than weaker programs when you double the cpu power. Weak
programs generally earn 70-80 points, compared to about 30-50 points for
stronger programs (given that the previous rating is around 2.000 or a few
hundred points more). There exists formulas for calculating the increase
in strength when you go from one CPU-speed to another (they are not
exact though).
> This is due to the exponential growth of
> the tree of possible moves. This is also the reason why Fritz 3
> seemingly plays worse in non-Blitz games: The program doesn't get much
> stronger with the extra time given while it's human opponents play much
> better if they're given enough time since humans don't try to play chess
> by using computation trees but with "true" intelligence ;-) (this is of
> couse an oversimplification of the problem ...)
>
> As always, ELO ratings are international FIDE, for American USCF ratings
> add 100.
>
> If you have trouble finding the evaluation function: It's only
> available if you play with fixed time limits and in "real
> tournament"-mode. This mode can be invoked by pressing the p key.
>
> I'd like to learn some "real" results from people who actually have
> Pentiums (or slower computers) and see if my approximation holds.
>
Wait for the next SSDF tournament-list and blitz-list, perhaps
Fritz 3 (and Genius 3.0!) will appear. I can mail them both if you want.
Thanks for an interesting post!
Greetings
Mads Brevik
Mark Chess
even 1900's when thinking 3 min. per move!
That seems like it has some major flaws. I don't see this happening with
MChess Pro, for example. It seems to me that it plays atleast master
strength at 3 min. per move.
A poem I just made up:
There was a computer named Fritz
That could only play good chess at blitz;
So if Kasparov played it,
He surely would mate it;
Only Chess Genuis 2 gives him fits!
Now you can see why I never write poems!!!!
Mark
Vincent Diepeveen
>> CPU benchmarks with Norton sysinfo 7.0 and PC Tools 9.0 si showed
>> the non-"turbo" mode to be approx. 1/3 as fast as "turbo" mode.
>>
>> On a Pentium 90 the program should (optimistically) have about 2540
>> int. FIDE ELO when playing 5 min. Blitz.
>
>I don't think 2.540 ELO is optimistically for Fritz 3 running on a
>Pentium 90. My guess is 2.550 on a 486/66 and 2.650 on a Pentium 90.
>Genius 2.0 (currently the strongest blitz-program rated by SSDF)
>has a blitz rating of 2.630 ELO (!) on a 486/66.
2550?
That is almost grandmaster level!
is 2300 not a better guess?
Only people who make huge blunders loose.
>> On the other hand, in regular
>> games (40 moves/120 minutes) playing strength should be about 2157 on
>> the same machine.
>>
>If you add 200 points I agree.. :)
Is your elo 2357 and do you even loose against fritzX??
What about positional play, playing solid, letting fritzX making
its own choices (where to put his rooks etc) and after having
developed yourself (of course you didn't play a line that was
in the openingsbook), it appears that fritz better can resign.
Of course don't give away pawns for space, that is stupid!
Don not give both player an open e-file. That is also stupid,
because this plays too simple for fritz(and other programms)
Give yourself an halfopen e or f file and give fritz an open c-file
(or visa versa). Computers know everything about little advantages,
but they don't know how to fine tune it: don't know de difference
between a knight on c4 or c5, it is near the centrum, NOT?
Playing the cool hazard(which kills you in the endgame if it sees
a good endgame), that is the winning advice.
First game(s) you loose then you win 100% and not because of the opening.
I win 90% of the games now against fritz2/Genius 2/CM4000.
I only sometimes play a draw against Genius 2 and CM4000, because
they find a tactical possibility which takes cares i loose a pawn,
and then draw the lost ending. All games were 10 minutes for the whole game,
and not looking to a 3-d chessboard is really a BIG handicap.
My national rating is something like 2200 (dutch),
and my fiderating something like 2300.
Fritz 2550, don't make me laugh!
>> I should say that the program doesn't take into account the size of hash
>> tables, although it shows a warning that you won't obtain full playing
>> strength when you start it with only 64KB hash tables (standard). It also
>> neglects more subtle details of hardware speed such as memory speed and
>> size and speed of the 2nd level cache.
>>
>> Since I played 60 games in 5 min. Blitz CM 4000 vs. Fritz 3, I tend to
>
>CM 4000 and Fritz 2 play almost at the same level in blitz. Fritz 2 is
>perhaps i bit sharper. If I don't remember wrong, CM 4K has about 2.450 ELO
>on a 486/66.
>> believe that the programs evaluation of its own playing strength should
>> be a good hint of its "real" strength, maybe with a correction of -50
>> ELO. This evaluation is of course true only with todays hardware (up to
>> Pentium), so nobody should try to calculate with the above rules of
>> thumb what hardware speeds are necessary to obtain a 3000 ELO with
3000, your banksaldo?
> Greetings
> Mads Brevik
Vincent Diepeveen
vdie...@cs.ruu.nl
--
+--------------------------------------+
|| email : vdie...@cs.ruu.nl ||
|| fidonet: 2:280/206.23 ||
+======================================+
Mads Brevik
> Mr. Berger pointed out that Fritz 3's rating decreased from 2400's down to
> even 1900's when thinking 3 min. per move!
Read the previous post for my personal views on these figures.
Usually a chess program 'loses' about 200 points when playing
at tournament-level compared to blitz level. Fritz 2 and CG 2.0 are two exceptions.
CG 2.0 plays well at both levels, while Fritz 2 doesn't do so well at tournament
level. They both play 300 points better at blitz-level (according to SSDF).
> That seems like it has some major flaws. I don't see this happening with
> MChess Pro, for example. It seems to me that it plays atleast master
> strength at 3 min. per move.
> A poem I just made up:
> There was a computer named Fritz
> That could only play good chess at blitz;
> So if Kasparov played it,
> He surely would mate it;
> Only Chess Genuis 2 gives him fits!
Nice poem!
> Now you can see why I never write poems!!!!
> Mark
Greetings
Mads Brevik
Mads Brevik
Subject: Re: Fritz 3 ELO on different computers
Newsgroups: rec.games.chess
Organization: Kongsberg Ingeniorhogskole
In article <CwH9r...@cs.ruu.nl> you wrote:
> In <35mkkt$k...@ratatosk.uninett.no> ma...@bacchus.kih.no (Mads Brevik) writes:
> >I don't think 2.540 ELO is optimistically for Fritz 3 running on a
> >Pentium 90. My guess is 2.550 on a 486/66 and 2.650 on a Pentium 90.
> >Genius 2.0 (currently the strongest blitz-program rated by SSDF)
> >has a blitz rating of 2.630 ELO (!) on a 486/66.
> 2550?
> That is almost grandmaster level!
> is 2300 not a better guess?
> Only people who make huge blunders loose.
Please notice that I'm talking about BLITZ-rating here (it should be
apparent if you read the article carefully)..
Do you really think that Fritz 3 Pentium 90 plays around 2.300 ELO
at blitz level, when Fritz 2 has 2.450 ELO on a 486/33?! (SSDF)
If Fritz 3/P90 played at 2.300 level do you think Kasparov would have that
much trouble beating the computer at blitz?
> >> On the other hand, in regular
> >> games (40 moves/120 minutes) playing strength should be about 2157 on
> >> the same machine.
> >>
> >If you add 200 points I agree.. :)
> Is your elo 2357 and do you even loose against fritzX??
No, my estimations are 'only' based on statistics for Fritz 2 (actual match
results , over 100) and what I know about Fritz 2. That is MUCH more reliable.
I don't stand any chance against Fritz 2 on my 486/66 at blitz-level.
> What about positional play, playing solid, letting fritzX making
> its own choices (where to put his rooks etc) and after having
> developed yourself (of course you didn't play a line that was
> in the openingsbook), it appears that fritz better can resign.
Fritz 2 is not a good positional player, that's true. That's
why the program plays 300-points weaker at tournament level.
> Of course don't give away pawns for space, that is stupid!
> Don not give both player an open e-file. That is also stupid,
> because this plays too simple for fritz(and other programms)
> Give yourself an halfopen e or f file and give fritz an open c-file
> (or visa versa). Computers know everything about little advantages,
> but they don't know how to fine tune it: don't know de difference
> between a knight on c4 or c5, it is near the centrum, NOT?
> Playing the cool hazard(which kills you in the endgame if it sees
> a good endgame), that is the winning advice.
Thanks for the advices, I know some of the 'tricks' :)
> First game(s) you loose then you win 100% and not because of the opening.
> I win 90% of the games now against fritz2/Genius 2/CM4000.
> I only sometimes play a draw against Genius 2 and CM4000, because
> they find a tactical possibility which takes cares i loose a pawn,
> and then draw the lost ending. All games were 10 minutes for the whole game,
> and not looking to a 3-d chessboard is really a BIG handicap.
What computer are you running your programs on? Not a 386 I guess? :)
Usually one can 'reduce' the computer's playing strength by around 100-150
points if you know it well. You're comparing yourself with the programs
when playing at 10 min/game, and that's not blitz-level. What are your
statistics against those machines at blitz-level? That's far more interesting
to me.
> My national rating is something like 2200 (dutch),
> and my fiderating something like 2300.
> Fritz 2550, don't make me laugh!
Tell me, have you tried playing Fritz 3 on a Pentium 90 at blitz? No?
A lot of grandmasters have (for not too long ago during an Intel-blitz tournament).
I don't think they would be so surprised by my 2.650 ELO estimate for Fritz 3/P90.
What about Fritz 2 486/66 then? If you're a good blitz-player (and know
some anti-computer tricks) you should win about 50% of the games, since
Fritz 2/Blitz-level plays slightly above 2.400 ELO on a 486/66 (2.400-100
vs 2.300). It can't be any fun to win 90% of the games, or is it?
Greetings
Mads Brevik