Hi,
I have a sony laptop running on windows 7 64 bit with an intel core i5 cpu m520 @ 2.4 GHZ with 4 gb ram. I am using the shredder classic GUI and running the SSE42 engine. I am new to the world of chess engines and would like some advise on how to maximise the intelligence of the chess program on my machine so it is playing at its strongest. Could someone tell me what I should be setting the hash tables at for optimal performance? Should I consider using the triple brain function or is this not as good? I have 3 engines to use, deep rybka 4, deep rybka sse42 and deep rybka 960 as well as my shredder chess engines. Which of these engines is the best?
Any help would be much appreciated as I am a newbie in this department.
Many Thanks
I have a sony laptop running on windows 7 64 bit with an intel core i5 cpu m520 @ 2.4 GHZ with 4 gb ram. I am using the shredder classic GUI and running the SSE42 engine. I am new to the world of chess engines and would like some advise on how to maximise the intelligence of the chess program on my machine so it is playing at its strongest. Could someone tell me what I should be setting the hash tables at for optimal performance? Should I consider using the triple brain function or is this not as good? I have 3 engines to use, deep rybka 4, deep rybka sse42 and deep rybka 960 as well as my shredder chess engines. Which of these engines is the best?
Any help would be much appreciated as I am a newbie in this department.
Many Thanks
Given that the total amount of RAM on your laptop is 4 Gb, I guess the maximum RAM you should use for Rybka is 1 Gb (more hash = better performance, but doubling RAM in this range of sizes achieves only a few percent of speedup). All of the above engines are the same, and for normal chess you should use either DR4 or DR4 SSE42 (this one is for new processors which use SSE42 set of instructions, and I guess yours qualifies, so use SSE42 verson, which will make better use of processors and be faster than ordinary DR4 by say 5% or so). DR4 960 is for special chess where initial position is decided randomly (except for pawns), but as for the rest it is the same as DR4. I'm not familiar with Triple Brain function in Shredder GUI but I guess it analyses the same position with 3 different engines (provided you have 3). This can give you an interesting extra view on the position, but will reduce the total time used by DR4 (believed to be the best) to one third of the total, so it is a bit of a tradeoff.
I have a I7 quad core machine with 6gb RAM, and uses SSE42 on default settings. It operates between 200 to 300 kn/s, and has never lost a ICCF game. So . . . there you have it
i'd recommend simply using the cpu/ram usage gadget (right click on desktop & click gadgets) & see what % of RAM your OS is using. close all obvious unneccsary processes and/or download gamebooster (do a quick google search) which will do this for you & free up some RAM. unfortunately sidebar (which the RAM meter gadget runs on) is one of these processes, but after you've checked to see what the RAM situation is you can close it with either task manager (ctrl+shift+esc) or gamebooster. the reason maximizing RAM is important isn't so much do to with hash size as with large pages, which is a new feature in rybka 4. it'll speed up the engine by a noticeable degree. i don't use it anymore myself, because -- and this important -- if you should switch engines, or close rybka, the large pages will not function without a reboot. to enable large pages just go the UCI settings & tick the box next to it. the reason RAM matters here is that there isn't much difference b/w 1 & 2GB in terms of hash size (afaik) but the larger the hash, the more benefit you will get from the large page function. i am not sure if a 4GB machine running win 7 is capable of using 2GB hash without slowdown...if not, it's no biggie, but if you can, why not?
i would also not recommend using triple brain or other engines if you are interested only in pure performance. however shredder is a very different engine from rybka (much slower & more knowledge based) & can at times provide some interesting alternatives. but it is at least 100 elo weaker.
i would also not recommend using triple brain or other engines if you are interested only in pure performance. however shredder is a very different engine from rybka (much slower & more knowledge based) & can at times provide some interesting alternatives. but it is at least 100 elo weaker.
There are optimums. The relevant factors are how much time is spent in a position on average and the kn/s rate of your machine. Usually it is between 16MB and 256MB. If it sits all day on one position then 512MB. A weekend or more: 1GB.
The only way to find your optimums is to do a lot of tests. The differences are not huge. 64 MB or 128MB are fine likely.
The only way to find your optimums is to do a lot of tests. The differences are not huge. 64 MB or 128MB are fine likely.
again, for large page usage i think this is incorrect. someone can correct me if i'm wrong, but afaik bigger is better. other than that, hash size is not a major issue with rybka (again, afaik).
Yes, bigger is better. Doubling hash size adds very roughly 5 Elo points unless you play really short games.
> If it sits all day on one position then 512MB. A weekend or more: 1GB.
















Why do people who have no idea on the subject feel compelled to say something???
Nowhere in the post did it say he was using large pages. Why are people compelled to belittle? I have run hundreds of thousands of games. No, I have not seen the code like some very select people, so I did not know that when running large pages the optimums are different. Good grief!
I have done extensive testing searching for optimum hash settings, though, admittedly most of that was with Rybka 3.
I have done extensive testing searching for optimum hash settings, though, admittedly most of that was with Rybka 3.
I am not speaking of large pages.
I just want to state that your "advice":
If it sits all day on one position then 512MB. A weekend or more: 1GB.
is rubbish (has no foundation)
I just want to state that your "advice":
If it sits all day on one position then 512MB. A weekend or more: 1GB.
is rubbish (has no foundation)
Even 1GB of hash would get filled up very quickly with today's processors. I am not sure how long it would take to fill up the hash but most likely not more than a couple of minutes and for sure less than an hour. For game play, where each move is played rather quickly, even in a LTC game 512mb might be plenty ... but for long term analysis the more memory the better, as for sure the hash will fill it up. As for choosing one of the R4 engines ... they basically are all the same but of course if your CPU supports SSE2 then go for that one as it will be around 5% faster. The 960 should evaluate the same as the default R4.
Filling the hash is a good thing. The bad thing is when you have to get rid of a potentially important position to allow storage of another potentially important position. With a larger hash, the loss of important positions should happen less frequently...
Even worse would be that your relatively fast RAM gets filled up and your hash is then saved on HDD ... not a pretty thought :)
If my opinion is based on experiment and is rubbish, I wonder what that makes your opinion.
> I wonder what that makes your opinion
See Majd's answer...
Hashtable filling/replacements is a matter of minutes. The hashtable couldn't care less about what happens after 24 or 48 hours...
There are housekeeping chores like decisions as to what to overwrite and the need to find a particular transposed position in a larger set of positions. Then we have to consider the fraction of positions recorded that will be of use in the future. In endgame that goes up but in the opening and middlegame there is going to be a lot of stuff written that will never be of any use. The sheer number of positions involved can begin to have an effect. At least that is my hypotheses. Everything is not a mater of reason...our imaginations are so limited...that is why the ancient Greeks did not figure out the world...it took experimentation to begin to have any real understanding of the world.
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill