After over 2 Years since release of i7-980x, if we look at the comparable latest version of today, the i7-3960, the improvement has been not more than 30% for chess.
There was a law in the past that every 18 months, speeds improve by 100%. This trend is no longer followed.
Since I didn't want to spend tons of money on a Xeon setup, I was expecting to see something fast coming up, but it seems that this good old i7-980x is still top notch for chess.
Less than two years ago (July 2010) the two top engines were Houdini 1.03 and Rybka 4, at IPON level around 2950.
Later this year Houdini 3 will be released, the current development version is about 110 Elo stronger than the above-mentioned programs.
This is equivalent to the strength increase in going from 1 to 8 cores.
In other words: Houdini 3 on 1 core will be at least as strong as Houdini 1.03 or Rybka 4 on 8 cores.
Basically your computer will have become 8 times more powerful in 2 years time.
Anyhow, I do also see a slowdown on that end since Houdini 1.5a no real progress was made. But I have to agree that the soft world is more "volatile".
> Maybe instead of giving our money into hardware company, we should give it to the programmers which make more/faster progress.
Of course, it's by far the cheapest way to give a massive boost to your hardware.
Every 40 Elo jump is the equivalent of doubling the number of cores of your hardware.
> Anyhow, I do also see a slowdown on that end since Houdini 1.5a no real progress was made. But I have to agree that the soft world is more "volatile".
A relative slowdown is quite normal after the huge jump of Houdini 1.5a that occurred only 16 months ago (December 2010).
Houdini 3 will be another huge jump, later this year.
You can either sit at the sideline complaining that progress is slow, or make a real contribution by supporting financially the engine authors that deliver.
We did already thought the same about version 2 but didn't materialized.
As for Houdini 3, see the Houdini Facebook page http://www.facebook.com/pages/Houdini-Chess-Engine/164560926948947 .
If there is already 40 ELO improvement over version 2, lets hope a new release is shortly so I can support it again.
For me 50 ELO is enough to be a new version of the programm to be purchased.
> We did already thought the same about version 2 but didn't materialized.
Depends how the engines were tested. When optimised with LP and appropriate split depth setting there is very evident improvement in Houdini 2.0c to 1.5a with matches played on 2 PC's so engine gets full resources of PC and ponder on. Testing with various books does not seem the best option these days when drawish lines in the book can diminish and mask the true performance of an engine. Fixed openings DB's such as the Noomen2012 for example are ideal because it ensures both engines have the opportunity to show their metal and reduces the need to play a mountain of games for the nth degree of Elo certainty.
Using fixed line DB's and the same group of opponents for H1.5a and H2.0c gives as accurate as possible comparison of relative engine performance. For my quads it works out that H2.0c is approximately 40 Elo better than 1.5a so when considering how much H1.5a was a cut above the rest then 40 Elo gain for 2.0c is a very significant and worthwhile at these levels.
> After over 2 Years since release of i7-980x, if we look at the comparable latest version of today, the i7-3960, the improvement has been not more than 30% for chess.
From 980x to 3960, why do you say 30%?
same number of cores, practically same frequency...
where would the improvement come from?
For chess?... what is involved?
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill