Chess2u
Would you like to react to this message? Create an account in a few clicks or log in to continue.

Chess2uLog in

descriptionchess tourney on a pragmatic notion (thought provoking essay) Emptychess tourney on a pragmatic notion (thought provoking essay)

more_horiz



many ppl run a chess tourney for fun
but what could be the best settings (mostly regarding TC)
for gaining the best and optimal results

to cut to the chase
the model answer could be something like
it depends on each person
and that could be the best answer!

that's the reason why this essay is thought provoking
since it doesn't have a definite answer
it makes us to think about this

if someone has a slow computer
with only a few cores
then he might want to opt for short TC
since his slow computer can't handle long TCs with deep searches

his computer is not up to par with running a long TC
so for him
a short TC is best for him

or he can even have single core tourneys only

I have a not so fast dual core CPU

I run 2 winboard 4.70 GUIs simultaneously
and 4 engines are loaded onto the RAM all the time
and 2 engines are in active use all the time
(ponder or permanent brain is off)

so I can let others with fast CPUs to run long TCs
while I only do short TCs

but if I run 2 single core tourneys at the same time
it is better since
I can pause one or two winboard GUIs
(so usually I can pause one GUI
and another GUI is running the tourney
so half of the CPU is used for the tourney
while another half of the CPU is used for other things)
when I have to use the computer
then when I don't use the computer
I can resume the tourney


***


if someone has a fast computer with many cores like 16
he can run one tourney with 16 cores
so an engine can use all of the 16 cores
so the search depth can be deep and

deeper search depth = higher quality matches

or he can run 16 single core tourneys simultaneously
so he can get 1600 single core matches every day
when he could only get 100 multicore matches a day

more matches = more accurate elo ratings

the person with a faster CPU has more options to choose from


***


there is no best answer for this essay question
but this reading can guide us to finding good answers
so after reading this essay
we can basically narrow down the possible good answers
before reading this essay
there could be 10 different good answers to the essay question
after reading this essay
we have less number of good answers
and that's a good way of finding the best answer!

if there are 10 good answers and we don't know what to pick
just remove the worst answers step by step
then we are only remained with a few good answers
then we can choose the best answer more easily!


***


if something (a certain overt setting and TC) suits you well
that is the best setting and TC for you
the answer to this essay could be subjective and not objective

so if a certain TC and tourney setting is good for you
then you can go ahead with that setting
a setting that works best for you may not be the best for others!


***


by the way
I'm running a new tourney for strelka 5.1
and SF Nov 5
even if not that matches are done for SF Nov 5
that version looks like a powerhouse
the initial elo result is super duper hyper strong

so far (if we have engine matches
between houdini and the lastest SF versions)
houdini is the best
when it comes to the matter of short TC and single core tourney

and SF is ahead of houdini (based on many other ppl's rating lists)
when it comes to long TCs
longer TCs (multicore)
and super long TCs

(houdini is the best in single core fast TC setting so far)
I think this SF Nov 5 may be the first chess engine
(or the first stockfish engine)
to win over houdini in single core tourney
(or single core fast TC)

this SF has an ample chance
to become the first chess engine ever in the world
to be ahead of houdini in single core fast TC
(so far single core tourney
and fast TC have been houdini's dominant territory!)

descriptionchess tourney on a pragmatic notion (thought provoking essay) EmptyRe: chess tourney on a pragmatic notion (thought provoking essay)

more_horiz



there are about 10 rating lists which are professional
and there are about other 10 rating lists which are done by individuals
which test SF dev versions everytime when the dev versions are released

of the 20 rating lists

SF is higher than H3 in elo
if long TC is used

H3 is higher than SF in elo
if rapid TC is used

***

from this site

http://ls-ratinglist.beepworld.de/

Rank Name Elo + - games score oppo. draws
1 Houdini 4 x64 3182 5 5 11000 66% 3065 42% (€)(best overall)
2 Houdini 3 x64 3150 3 2 56000 66% 3030 41% (€)
3 Robodini 1.1 x64 3143 5 5 11000 64% 3040 43% (sc)(sp)
4 Stockfish 131111 x64s 3121 4 4 14000 57% 3071 48% (best mp-free+open)

this is a favortie list for many ppl

***

this is a super thought provoking essay
that is like an armchair brainstorming and thought experiments

these are just my personal and subjective opinions
and nonobjective thoughts

in my opinion
it is not easy to say which TC is better

I'm of the opinion that
it all depends on ppl

if someone likes one TC better than other TCs
that's the answer
if somebody prefers certain TCs to others
then that's good for him

in my opinion
I like the rapid TC
actually I only think about rapid TCs
and the following is the subjective opinions and reasons why

when I play a chess match with com
or when I analyze a pgn or a match by using com
or when I let engines play a tourney
to have engine versus engine matches

I only use rapid TC mostly
cuz when I play a match with com
1 sec is enough for my need
if I can play 100 times better than I can now
then the engines need more time to think to make a move
but engines are superb
they only need 1 sec even on a very old computer

decades ago,
when we had 286, 386, 486, 586, pentium 1 and so on
the speed of CPU calculation was slower
so maybe engines needed long TCs
but we are now in the new age with super fast com

I think 1sec per move is more than enough for many ppl's need
cuz with 1 sec per move
the engines can find almost perfect moves instantly

also, if I set the TC long
it means
I'll play chess matches with com
hundreds of thousands of times in total
and for each and every move
if the computer calculates for 10 or 30 sec
then it means
30 sec * number of rounds in total
= I'll have to spend thousands of hours in total
only to sit and wait for nothing absent-mindedly

so, in my opinion
I only think about rapid TC rating lists
cuz rating lists with long TCs are just too much
I think they are surreal
I think they are like not realistic

cuz there are ppl like me
and to take an example of me
I only use rapid TC
I never use long TC at all
when I play a match with com
when I analyze a match
or for engine tourneys

100% of all the time
rapid TC is the only TC setting needed for me
so I just think about rapid TC
and never even think about long TC
I don't need long TC when I don't even use long TC

***

it is just like this analogy and comparison

if a person, A only has money for buying a cheap car
then A doesn't even have to think or think about expensive cars
since it is useless
cuz expensive cars are out of reach
and surreal and not realistic
it is only usable in dreams

also, we know that
there is almost unlimited amount of extremely expensive metal resources
on the moon
but we don't have to think about those resources
since they are far-fetched
and surreal and not realistic
even if there're that much resources on the moon
it's impossible to use them
the resources are useless since it is unreachable

likewise,
for many ppl I know and for me
ppl only use rapid TCs mostly or all the time
for those ppl,
they only value rapid TC tourney results
since long TC results are not even usable for them
and long TC results are far-fetched for them
also long TC results are unreachable and unrealistic
they are surreal
it is like talking about
"I'm a rich guy, and I just bought the super expensive tires
for my car that is one of the most expensive in the world"
to a poor person
who can barely own the cheapest used car

so, I only think about rapid TC rating lists
since I almost always never even have occasions to deal with long TC matches
why should I think about long TCs
when I never even use long TCs in reality realistically speaking

I only value the results of rapid TCs
and I don't even take long TC results into consideration

also I think if an engine is good in the rapid TC
then that's it
cuz I think the tourney result of rapid TC is of the most value
this is my opinion anyway

***

here is an example
why I think the tourney result of rapid TC is of the most value
and why I don't even think about the result of long TCs

in universities
ppl have tests and exams
and almost always in the entire world as far as I know and heard
mostly, the exam time is less than a few hours
usually 1 or 2 hours
and no more than 3 hours usually
there is a reason that is tangible and pragmatic
why we don't allow students to think for more than 2 hours

during that short amount of time
ppl have to show what they got
and what they can muster and think

so most ppl can barely finish answering for the exam just on time
if we increase the exam time from 2 to 4 hours
then ppl can slowly and lethargically answer
without making a hard effort or without concentrating severely

there are many reasons why the exam time is almost always short and very short
ppl barely have enough time (and usually not enough time) to answer on time
and not enough for the amount to be answered

one another reason is that
if students need exactly 2 hours to answer properly
then when 2 hours are given to students
then students can barely finish the exam in 2hours
then 2 hours is a short amount of time
(= rapid TC tourney)
but if 4 hours are given
then it is a common knowledge I learned
that the students' knowledge is not correctly measured if the exam time becomes long

so I think
with rapid TC settings
we can most accurately and correctly measure the engine's elo strength

***

there are many reasons but
one reason is
like I aforementioned in another post in another thread

some engines (A) use almost 100% of the given time
while there are many engines (B) which only use about 70% of the allocated time

if rapid TC is used
A uses 1.0 sec per move on average
while B uses 0.7 sec per move
but using 0.3 less sec makes a small difference
in terms of searched depths

but if long TC is used
A uses 1 hour per round
while B uses only 42 min
and using about 20 min less makes a super big difference
in terms of searched depths for every move
and some engines are fast thinkers
they only use 70% or so of the given time

***

that exam time was one real life example and analogy

there are many other real life examples
but here is one more

if I own a company
then making fast decisions and calculating for the future moves fast
is of utmost importance in business and economy

I'll only employ those subordinates
who can think and calculate fast during the given time

***

when I had computers many years ago
my computers were slow
so I sometimes (very rarely) played engines under long TCs
since using long TCs made a perfect sense

but now I have a faster computer
the situation was changed

(also new engines are written
in a much more efficient and effective ways
newer engines run faster
cuz the coding and code structure are better
also the newer compilers can make faster compiles)

so for my tourneys
when the situation is changed (CPUs became faster)
the TCs for my tourneys can be shorted
so rapid TC is the way to go
cuz it just sounds logical


Permissions in this forum:
You cannot reply to topics in this forum